Manifesting hidden dynamics of a sub-component dark matter Ayuki Kamada, 1, 2, 3, * Hee Jung Kim, 1, 4, † Jong-Chul Park, 5, ‡ and Seodong Shin 1, 6, § 1 Center for Theoretical Physics of the Universe, Institute for Basic Science (IBS), Daejeon 34126, Republic of Korea 2 Kavli Institute for the Physics and Mathematics of the Universe (WPI), The University of Tokyo Institutes for Advanced Study, The University of Tokyo, Kashiwa 277-8583, Japan 3 Institute of Theoretical Physics, Faculty of Physics, University of Warsaw, ul. Pasteura 5, PL–02–093 Warsaw, Poland 4 Department of Physics, KAIST, Daejeon 34141, Korea 5 Department of Physics and Institute of Quantum Systems (IQS), Chungnam National University, Daejeon 34134, Republic of Korea 6 Department of Physics, Jeonbuk National University, Jeonju, Jeonbuk 54896, Republic of Korea (Dated: November 15, 2021) We emphasize the distinctive cosmological dynamics in multi-component dark matter scenarios and its impact in probing a sub-dominant component of dark matter. We find that the thermal evolution of the sub-component dark matter is significantly affected by the sizable self-scattering that is naturally realized for sub-GeV masses. The required annihilation cross section for the sub-component sharply increases as we consider a smaller relative abundance fraction among the dark-matter species. Therefore, contrary to a naive expectation, it can be easier to detect the sub-component with smaller abundance fractions in direct/indirect-detection experiments and cos- mological observations. Combining with the current results of accelerator-based experiments, the abundance fractions smaller than 10% are strongly disfavored; we demonstrate this by taking a dark photon portal scenario as an example. Nevertheless, for the abundance fraction larger than 10 %, the warm dark matter constraints on the sub-dominant component can be complementary to the parameter space probed by accelerator-based experiments. * [email protected]† [email protected]‡ [email protected]§ [email protected]arXiv:2111.06808v1 [hep-ph] 12 Nov 2021
48
Embed
University of Warsaw, ul. Pasteura 5, PL{02{093 Warsaw ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Manifesting hidden dynamics of a sub-component dark matter
Evidences for the existence of dark matter (DM) come from observing the gravitational influence of
DM alone in various length scales of the Universe. On the other hand, the particle nature of DM is
elusive and our practical viewpoint on DM remains to be a bulk of matter that is dominant in mass.
In the last few decades, there have been extensive efforts to search for non-gravitational interactions
of DM with the Standard Model (SM) particles whose mass and interactions are set by the weak scale
and the weak interaction of the SM, i.e., the weakly interacting massive particles (WIMP). Mainly
due to the lack of any conclusive experimental signals of non-gravitational interactions of WIMP
so far [1, 2], many alternative scenarios of dark sector beyond WIMP have been proposed recently.
Among them, the scenarios of non-minimal particle contents inside a dark sector have drawn lots
of attention because of their abilities resolving various phenomenological issues and providing extra
power to many current/future experiments of searching for their signals in new and creative ways.
Examples include the scenarios of inelastic DM [3], self-interacting non-minimal dark sector to address
small-scale issues [4–10] and the existence of the supermassive blackholes at high redshifts [11–13],
and multi-component boosted dark matter (BDM) whose unique signals can be probed in a variety
of neutrino and direct-detection experiments [14–27]. Nevertheless, less attention has been given to
exploring the cosmological dynamics of the sub-dominant component of DM and the corresponding
impact on their detectability.
A sub-dominant component of DM can play a dominant role in the dynamics of a dark sector.
We already know an example in the SM. Electrons, a component of matter that is negligible in
mass compared to baryon, play an important role in coupling baryons with photons in the early
Universe. The observed baryon acoustic oscillations in the cosmic microwave background (CMB)
anisotropies imply that the baryon and the photon bath were tightly-coupled until the recombination
epoch. Without the help of electrons, protons, a dominant component of the baryon, cannot couple to
photons until then. Although electrons have negligible gravitational influence in the point of view from
a dark sector, they are actually dominant in interaction and play an important role in the cosmological
evolution of the baryon. This well-known example can be a motivation for paying attention to a
sub-component dark matter in a variety of dark-sector scenarios beyond WIMP.
Probes of a sub-dominant component in a dark sector can be promising when it has sizable inter-
actions with the SM particles. It is well known that a wide range of parameter space of a vanilla
model of Higgs portal DM, where the interactions in the thermal freeze-out and the direct-detection
experiments are essentially same up to the crossing symmetry, is strongly constrained even when the
DM is a sub-dominant component whose mass is & O(GeV) [28–31]. This is because the large cou-
pling between the DM and the SM particles, which is essential in suppressing its fraction in the total
DM, increases the scattering cross section between the DM and the target nucleus in direct-detection
experiments. Hence, the fraction of the sub-dominant component Ωsub/Ωdm,total entering linearly in
the direct-detection signal rate is canceled by the large coupling squared in the cross section, allowing
the experimental constraints to be applied to the sub-component DM equally.
The strategy to probe a sub-dominant DM component relies on its cosmological evolution, sensitive
to the interaction within a dark sector, as stated in the previous paragraph. In this paper, we study a
case where the dynamics within a dark sector affects the detectability of a sub-dominant component.
For a concrete demonstration, we take the minimal two-component DM scenario, where the relic
density of two stable DM components are determined by the assisted freeze-out [32]; the heavier
DM particle χ0, which is dominant in mass, is secluded from SM and directly annihilates only into
3
the lighter DM particle χ1 while the sub-dominant component χ1 annihilates into the SM particles.
We show that the dynamics of the assisted freeze-out entails larger annihilation cross sections of
the sub-dominant component compared to standard freeze-out scenarios. This renders the enhanced
detectability of χ1, e.g., in cosmological/astrophysical observations. We highlight the cosmological
evolution of χ1 by taking into account a large self-scattering cross section of χ1, i.e., σself/m ∼0.1 cm2/g, which is naturally realized for sub-GeV mass-scale of χ1 in our reference set-up. The
collaboration between the χ0-annihilation and the strong self-scattering among χ1 leads to a distinct
thermal evolution of χ1, which we dub as DM self-heating [7, 9, 33, 34]. The enhanced temperature of
χ1 from DM self-heating affects their velocity-dependent annihilation rate during cosmological epochs
sensitive to DM annihilation. Furthermore, the resultant warmness of χ1 from DM self-heating affects
their gravitational clustering and leaves imprints in matter power spectrum. In order to guide the
attention of readers to their own interests, we devote the rest of the section to providing a scope of
our analyses.
Scope of our analyses
The chemical freeze-out of the sub-dominant component χ1 has a distinctive feature from the stan-
dard freeze-out of single-component DM scenarios. If the relic abundance of χ1 is negligible around
its freeze-out, i.e., r1 = Ωχ1/Ωdm,tot 1, the production of χ1 from χ0-annihilation is non-negligible
around its freeze-out. Consequently, the required annihilation cross section of χ1 is sharply enhanced
towards smaller r1. In the case of s-wave (p-wave) annihilation, the required annihilation cross section
of χ1 scales as 1/r21 (1/r3
1), in contrast to a naive expectation, scaling as 1/r1. It is worthwhile to
note that considering smaller values of r1 is sometimes referred as a minimal remedy to evade the
stringent indirect-detection constraints on sub-GeV DM annihilations (for example, see Ref. [35]). We
remark that this is not entirely true because of the sharp enhancement of the χ1-annihilation rate
towards smaller r1 in our reference scenario. We provide semi-analytic understandings on the chemical
freeze-out of DM in the two-component DM scenario in Section II A. In order to focus on the impact
of the distinct dynamics of the chemical freeze-out, we review the cosmological/experimental bounds
on χ1 while turning-off the self-scattering of χ1 by hand in Section II B.
Moreover, we highlight the impact of the self-scattering among the lighter DM particle χ1 on its
cosmological evolution. After the freeze-out of all the DM particles, residual annihilation of the χ0
produce χ1 particles which have enough energy for self-heating due to the mass difference. This self-
heating enhances the temperature of χ1 and affects its observable signatures such as the suppression
of the gravitational clustering in the Galactic scale. Hence, the constraints for warm dark matter
(WDM) enter even for mχ1 O(keV) and the interpretations of the experimental/observational
results from DM direct-detection experiments and the diffuse X-ray/γ-ray background should be dif-
ferent. Furthermore, if the annihilation of χ1 is velocity-suppressed, DM self-heating enhances the
annihilation rate of χ1 during the cosmological epochs sensitive to DM annihilation, e.g., during the
photo-dissociation epoch [36] and at the last scattering. Consequently, the cosmological observations
can have more constraining power on the annihilation cross section of χ1. The thermal evolution of χ1
with its self-heating and its impact on cosmological/astrophysical signatures are discussed in Sec. III A
and Sec. III B, respectively.
Although the interesting cosmology for r1 1 provides new possibilities on detecting χ1 in cos-
mological observations, we remark that for abundance fractions smaller than r1 . 0.1, the enhanced
4
interaction between χ1 and SM is usually incompatible with the constraints from terrestrial experi-
ments. In Section IV, we demonstrate this argument for a reference model of two-component singlet
scalar DM with a dark photon mediator. We highlight that the WDM constraints from DM self-heating
can be complementary to the parameter space probed by terrestrial experiments.
We conclude in Section V. Further details on the Boltzmann equations of DM, and the temperature
evolution of χ1 in the presence of DM self-heating are collected in Appendix A, B, C and D.
II. COSMOLOGY OF TWO-COMPONENT DARK MATTER
A. Chemical freeze-out
In this section, we revisit the processes of the chemical freeze-out of DM particles in a simple reference
scenario, two-component DM (χ0 and χ1) with mass hierarchy mχ0 > mχ1 and the following processes:
• Annihilation of χ0: χ0 + χ0 ↔ χ1 + χ1.
• Annihilation of χ1: χ1 + χ1 ↔ sm + sm, where “sm” stands for Standard Model particles.
• Elastic scatterings: χ1 + χ0 → χ1 + χ0 and χ1 + sm→ χ1 + sm.
The DM particles are initially in thermal equilibrium with the SM plasma. During the decoupling
of the annihilations, we assume that DM are in kinetic equilibrium with the SM plasma. This is
justified by the crossing symmetry between DM annihilations and DM elastic scatterings; the rate
of elastic scatterings of a DM particle with some lighter state is typically larger than that of DM
annihilations by a factor of ∼ nlight/ndm. For χ0, the rate of χ0 + χ1 ↔ χ0 + χ1 is larger than the
rate of χ0 + χ0 ↔ χ1 + χ1 by a factor of ∼ nχ1/nχ0 and hence decouples later; the similar discussion
works for χ1. For convenience, we introduce the DM yield, Yχi = nχi/s, in addition to x = mχ1/T ,
where s = (2π2/45)g∗ST 3 and g∗S is the effective number of relativistic degrees of freedom in entropy
density. Assuming the kinetic equilibrium, the evolution equations for the DM yields are [32]
dYχ0
dx= −λχ0
(x)
x
[Y 2χ0−(Y eqχ0
(x)
Y eqχ1 (x)
)2
Y 2χ1
],
dYχ1
dx= −λχ1
(x)
x
[Y 2χ1−(Y eqχ1
(x))2]
+λχ0
(x)
x
[Y 2χ0−(Y eqχ0
(x)
Y eqχ1 (x)
)2
Y 2χ1
],
(1)
where we have defined the dimensionless rates λχi = s 〈σivrel〉 /H, and H2 = g∗π2T 4/(90m2pl) with
mpl being the reduced Planck mass. The thermally averaged annihilation cross section 〈σ0vrel〉 is for
the χ0χ0 → χ1χ1 while 〈σ1vrel〉 is χ1χ1 to SM particles. In this paper, we explicitly show the velocity
dependence as 〈σivrel〉 ' (σivrel)s + (σivrel)p〈v2rel〉. For simplicity, we focus on the regime where the
annihilation of the heavy component χ0 decouples first while the lighter component χ1 remains in
thermal equilibrium. In such a case the chemical freeze-out of χ0 proceeds like the standard WIMP
freeze-out, and the asymptotic value of the yield is
Yχ0(∞) ≈ n0 + 1
λχ0(xfo,0)
, (2)
5
where n0 = 0 in the case of s-wave annihilation of χ0 and xfo,0 = mχ1/Tfo,0 with Tfo,0 ∼ mχ0
/20
being the freeze-out temperature. 1 Note that the estimation of Yχ0(∞) can considerably change for
mass difference as small as δm = mχ0 − mχ1 . mχ0/10 where the chemical freeze-out processes of
χ0 and χ1 interfere. Even for δm & mχ0/10, the interference occurs in the case that χ0 abundance
is exponentially suppressed, i.e., r0 = 1 − r1 e−δm/Tfo,0 , since the freeze-out of χ0 can be delayed
and thus interfere with that of χ1. 2 Hereafter, we will implicitly avoid such regime and focus on the
simplest case where χ0 relic abundance is estimated as Eq. (2), as our main purpose is to demonstrate
the impact of self-heating in a given scenario.
The estimation of the final yield of χ1 is more involved. After the χ0 freeze-out, evolution of Yχ1is
written as
dYχ1
dx' −λχ1
(x)
x
[Y 2χ1−(Y eqχ1
(x))2 − Y 2
ast. (x)], (3)
where Yast. is defined as
Yast. (x) =
√〈σ0vrel〉〈σ1vrel〉
Yχ0(x) . (4)
The term proportional to Yast. represents the light DM production from the heavy DM annihilation,
χ0χ0 → χ1χ1. If Yast. is negligible compared to Y eqχ1
around the standard freeze-out point of χ1, i.e.,
Tfo,1 ∼ mχ1/20, the final relic of χ1 is estimated as Yχ1(∞) ≈ (n1 + 1)/λχ1(xfo,1) where n1 = 0 in
the case of s-wave annihilation of χ1 and xfo,1 = mχ1/Tfo,1 with Tfo,1 ∼ mχ1
/20 being the freeze-out
temperature. But as we consider smaller r1 1, xfo,1 would become larger while Y eqχ1
(xfo,1) ∝ e−xfo,1
becomes more suppressed; eventually, the production rate of χ1 from the χ0-annihilation becomes non-
negligible compared to the annihilation rate of χ1 into the SM particles where we dub this situation
assisted regime. In the assisted regime, the final relic would be larger than the estimation in the
standard freeze-out regime. Below, we discuss the estimation of the final yield in the two illustrative
cases, i.e., the cases of s-wave and p-wave annihilation of χ1 while the χ0-annihilation is fixed to
be s-wave for simplicity. Nevertheless, the analytic estimations we present can be used for general
partial-wave annihilations of DM.
Figure 1 shows the numerical solutions to Eqs. (1) in the case that the χ1-annihilation is s-wave. The
left (right) panel shows the chemical freeze-out in the assisted (standard) freeze-out regime. We also
present the solution in the case of standard freeze-out, i.e., ignoring Yast. in Eq. (3), as the thin solid line.
We dub this standard regime for simplicity. In the assisted regime, the final yield of χ1 is significantly
enhanced compared to the case of standard regime. Around x ∼ 30, instead of following the equilibrium
trajectory (dotted) further, Yχ1 follows the constant Yast. (purple) asymptotically; this is because
the volumetric production/annihilation rate from χ0-annihilation/χ1-annihilation balance there, and
hence the yield of χ1 seizes to decrease down to the yield predicted in the case of standard freeze-
out. The final yield of χ1 is estimated by the balance condition as Yχ1(∞) ≈ Yast.(∞). The detailed
analytic arguments for this estimation can be found in Appendix A. Putting our understandings in
the standard/assisted freeze-out regimes together, we estimate the final yield of χ1 as
Yχ1(∞) ≈ max
[Yast.(∞),
n1 + 1
λχ1(xfo,1)
], (5)
1 We determine Tfo,0 as in the case of freeze-out of WIMP, following Ref. [37].2 The chemical freeze-out with small mass differences and exponentially suppressed r0’s are explored in Refs. [38, 39].
FIG. 2. Same as in Figure 1, but in the case of p-wave annihilation of χ1. In the left panel, the departure
point of Yχ1 from Yast. is x′fo ' 79 [Eq. (9)].
where Yast.(∞) is given as
Yast.(∞) =
√(σ0vrel)s(σ1vrel)s
Yχ0(∞) . (6)
When the first (second) term inside the maximum determines the final yield of χ1, the freeze-out of
χ1 is in the assisted (standard) regime. Note that (σivrel)s (and (σivrel)p later) is independent of the
velocity vrel in our notation. In the assisted freeze-out regime, the required annihilation cross section
of χ1 for a given r1 is
(σ1vrel)s ' 4.7× 10−24cm3/s
(0.1
r1
)2(mχ1
/mχ0
0.6
)2(√g∗g∗S
)xfo,0
. (7)
We remark that the annihilation cross section is enhanced towards smaller values of r1 as (σ1vrel)s ∝1/r2
1. The r1-dependence of the annihilation cross section in the assisted regime is sharper than that
in the standard freeze-out regime where the annihilation cross section scales as ∝ 1/r1.
7
Figure 2 shows the numerical solutions to Eqs. (1) in the case of p-wave annihilation of χ1 pair into
the SM particles. The left (right) panel shows the chemical freeze-out in the assisted (standard) freeze-
out regime. Again, the final yield of χ1 in the assisted freeze-out regime is significantly larger than
what is expected in the standard freeze-out regime, as clearly seen by comparing the thick and thin blue
curves in the left panel of Figure 2. The difference from the case of s-wave annihilation of χ1 is that
Yχ1follows Yast. (purple) only until x ∼ 100 and gradually reaches a constant value asymptotically
since the asymptotic value of Yast.(∞) is no longer a (constant) scaled value of Yχ0(∞); the ratio
〈σ0vrel〉/〈σ1vrel〉 now increases as the temperature T decreases. We denote the SM temperature at the
departure point from Yast. as T ′fo. The final yield of χ1 roughly coincides with Yast.(x′fo). More precisely,
the final relic abundance χ1 in the assisted regime can be estimated as Yχ1(∞) ≈ (n1 + 1)/λχ1(x′fo)
where the x′fo = mχ1/T ′fo is defined by the point where the relative deviation of Yχ1
from Yast. becomes
order unity; detailed analysis and the accuracy of this estimation are collected in the Appendix A. We
estimate the final yield of χ1 as
Yχ1(∞) ≈ max
[n1 + 1
λχ1(x′fo)
,n1 + 1
λχ1(xfo,1)
]. (8)
When the first (second) term inside the maximum determines the final yield of χ1, the freeze-out of χ1
is in the assisted (standard) regime. At x = x′fo, (Yast. − Yχ1)/Yast. = c′ and c′ ' 0.35 is a numerical
constant to fit the final relic abundance to numerical results. x′fo is given by
x′fo ' 47
(c′
0.35
)2/3(mχ1/mχ0
0.6
)2/3((σ1vrel)p
4.5× 10−23 cm3/s
)1/3(g∗S√g∗
)2/3
x′fo
(√g∗
g∗S
)1/3
xfo,0
. (9)
The required annihilation cross section of χ1 for a given r1 is given as
(σ1vrel)p ' 4.2× 10−24 cm3/s
(c′
0.35
)4(mχ1/mχ0
0.6
)4(0.1
r1
)3(g∗S√g∗
)4
x′fo
(√g∗
g∗S
)2
xfo,0
, (10)
where we define (σ1vrel)p through the relation 〈σ1vrel〉 = (σ1vrel)p〈v2rel〉 with the thermal average of
the squared relative scattering velocity among χ1, 〈v2rel〉 ' 6Tχ1
/mχ1. One would recover the value of
(σ1vrel)p used in Eq. (9) by taking g∗ = g∗S = 10.75. Note that the r1-dependence of the annihilation
cross section, i.e., (σ1vrel)p ∝ 1/r31, is even sharper than in the case of s-wave annihilation cross section
(σ1vrel)s ∝ 1/r21. This is because Yast.(x) increases with x contrary to the s-wave case, due to the
velocity dependence of 〈σ1vrel〉 for the p-wave case [Eq. (4)]. Therefore, Yχ1 is lifted up more by
following Yast. until x ∼ x′fo. From Eq. (9), the value of x′fo increases for larger values of (σ1vrel)p,
keeping the above effect longer. We remark that, regardless of the DM masses, the assisted regime
emerges as we consider r1 1, i.e., the first term inside the maximum dominates over the second
term in Eqs. (8) and (5) for r1 1. This is because in the assisted regime, the required annihilation
cross section to realize a given r1 exhibits sharper dependence on r1, (σ1vrel)s,p ∝ 1/r2,31 , compared to
the standard freeze-out regime, ∝ 1/r1. The sharp enhancement of the required cross section towards
smaller r1 generally makes the two-component DM scenario tightly constrained by direct-detection
experiments and cosmological observations compared to the single component DM case, as will be
discussed in the next section.
B. Scenario without DM self-heating
After the chemical freeze-out of DM, residual χ1-annihilations can produce significant flux of en-
ergetic SM particles that can be probed through cosmological/astrophysical observations. The non-
8
observation of such signatures provides bounds on DM annihilation cross sections. Meanwhile, if χ1
exhibit sizable self-scattering, the temperature evolution of χ1 could be sensitive to χ0-annihilations;
the residual χ0-annihilations may lead to DM self-heating. The modifications on the thermal history
of χ1 could directly affect the bounds on χ1-annihilation if the annihilation cross section of χ1 depends
on Tχ1 .
In this section, in order to focus on the impacts of introducing the assisted regime, we first review the
thermal history and the cosmological/experimental bounds on χ1 while turning-off the self-scattering
of χ1 by hand. Note that it is actually more natural to expect sizable self-scattering among χ1 particles
for our reference mass range of mχ1< O(0.1 GeV) in many multi-component dark matter scenarios,
which will be discussed in later sections.
1. s-wave annihilation of χ1
The cosmological/astrophysical bounds on DM annihilations are very stringent for sub-GeV DM
due to their enhanced number density. If DM dominantly annihilates into electromagnetic particles,
the bounds on sub-GeV DM annihilations disfavor the standard single-component thermal DM in the
case of s-wave annihilation. The two-component DM scenario is sometimes considered to be a minimal
remedy to be consistent with the stringent bounds on DM annihilations [35]; the sub-dominant DM
component χ1 with abundance fraction r1 1 annihilates into SM with the annihilation cross section
enhanced as (σ1vrel)s ∝ 1/r1 and the volumetric annihilation rate is suppressed towards smaller r1 as
n2χ1
(σ1vrel)s ∝ r1. Since the bounds on DM annihilations are basically given in terms of the quantity
proportional to the volumetric rate, considering r1 1 seems to be a viable possibility at the first
sight. However, this is not entirely true since, as we have seen in Section II A, the relic abundance
of χ1 is determined in the assisted regime where the required annihilation cross section scales as
(σ1vrel)s ∝ 1/r21 and thus the volumetric annihilation rate is virtually independent of r1. Therefore,
considering smaller r1 may not relax the bounds on χ1 annihilation as naively expected. In the rest of
this section, we review the possible indirect-detection constraints on s-wave χ1-annihilation for a vast
range of the abundance ratio r1, keeping in mind the caveat on the required annihilation cross section
in the assisted regime. Since the constraints on s-wave annihilation do not depend on the temperature
evolution of χ1, we leave the discussion on the temperature evolution for the next section where we
discuss the case of p-wave annihilation of χ1. We first summarize the considered list of constraints
below.
• Bounds on MeV-scale freeze-out of DM [40]: Light DM particles that are in thermal equi-
librium exclusively with the baryon-photon plasma or neutrinos, beyond the neutrino decou-
pling, i.e., T . Tν,dec ∼ 2 MeV, are constrained by the BBN and CMB observations. Around
T ∼ 1 MeV, DM energy density may considerably contribute to the expansion rate of the Uni-
verse, and DM annihilations may release significant amount entropy exclusively into the baryon-
photon plasma or neutrinos. As a consequence, the temperature ratio between neutrinos and
photons after the neutrino decoupling and the synthesis of the primordial elements during BBN
may be considerably affected. Cosmological observables such as Neff from the CMB observa-
tions and the observations on the primordial abundances of light elements (e.g., helium and
deuterium) from BBN will thus provide constraints on the DM mass. DM with masses greater
than mdm & 40 MeV will not be constrained since they freeze-out before the neutrino decou-
pling and their energy density is negligible around T ∼ 1 MeV. For electrophilic thermal DM,
9
DM annihilations will raise the photon temperature and thus lead to Neff smaller than the SM
prediction. We will adopt the constraint coming from the Planck data alone [41], rather than a
joint analysis with both the local measurements and BBN observations. 3 For a complex scalar
DM, which will be the illustrative case in Section IV, Planck data alone constrains DM mass to
be mdm & 4.6 MeV at 95.4 % CL [40]; remark that the constraints apply irrespective of r1.
• Photo-dissociation constraints on DM annihilation [36]: The residual annihilation of DM
after the freeze-out could affect the abundances of light elements through the process of photo-
dissociation. We first briefly review the case of DM mass larger than a few GeV [45], and discuss
the caveats of sub-GeV DM annihilations. When DM annihilate into electromagnetic components
of SM, e.g., e+e− or γγ, the energetic final state particles initiate the electromagnetic cascade,
e.g., by scattering with background photons, thermal electrons, and nuclei. The cascade process
redistributes the injected energy from DM annihilations among the electromagnetic particles and
produces an energetic photon spectrum. The photon spectrum is exponentially suppressed above
a high-energy cutoff, E ∼ m2e/22T ; photons above the cutoff are efficiently degraded through
the pair annihilation process γγb → e+e− (γb denotes the background photon) [46–48]. When
the cutoff is larger than the thresholds of the photo-dissociation processes of light nuclei, e.g.,
D, 3He and 4He, the processes are triggered. The triggered photo-dissociation processes modify
the abundance ratios among the light nuclei. The predicted abundance ratios are compared
with the observed values to give upper bounds on DM annihilation cross section. The photo-
dissociation processes from DM annihilation are relevant at temperatures long after the BBN,
100 eV . T . 10 keV [36, 49]; for T . 10 keV, the high-energy cutoff of the resultant photon
spectrum become larger than the dissociation thresholds of light nuclei; for T . 100 eV, the
high-energy cutoff is larger than the dissociation thresholds (of 4He) while the energy injection
rate redshifts towards lower temperatures.
For the annihilation of sub-GeV dark matter, small mdm renders the high-energy cutoff at E ∼min
[m2e/22T,mdm
]; this is because photons of E & mdm are limited by the initial energy
injection spectrum from DM annihilation. 4 For example, for mdm . 2 MeV, the cutoff is smaller
than the threshold energy of D and the photo-dissociation constraint disappears. We employ the
photo-dissociation constraints on sub-GeV DM annihilations presented in Ref. [36]. For s-wave
annihilating χ1, we simply rescale the constraints with respect to the factor r−21 .
• CMB bounds on DM annihilation [41]: After the freeze-out of χ1, residual annihilation of
χ1 into SM particles continues all the way down to the recombination epoch. Although their
annihilation rate per volume, ∼ n2χ1〈σannvrel〉, is tiny, it can be significant enough to affect the
CMB through energy injection into the SM plasma. The energy injection from DM annihila-
tion ionizes the neutral hydrogen and modifies the ionization history between the recombination
and the reionization. The additional free electrons scatter with CMB photons and make the
last scattering surface thicker. The broadening of the last scattering surface affects the CMB
3 Ref. [40] provides limits on DM mass through joint analyses by combining the Planck data with the local measurement
of H0 [42] (Planck+H0), or with the measurements of the primordial abundances of light nuclei [43] (Planck+BBN).
Each joint analysis prefers larger values of Neff compared to the analysis of the Planck data alone for non-annihilating
DM. This is because of the apparent tension on the determination of H0 from local measurements and Planck data,
and the slight ∼ 0.9σ tension on the inferred Ωbh2 between BBN and CMB observations [44]. Since electrophilic DM
lowers Neff , joint analyses provide stronger limits on the masses of electrophilic DM; for complex scalar DM, the limits
are mdm & 9.2 MeV for Planck+H0 and mdm & 8.1 MeV for Planck+BBN. To be conservative, we take the limit from
Planck data alone.4 See Refs. [36, 49, 50] for the dedicated analyses and discussions on the resultant photon spectrum when the high-energy
cutoff is limited by the DM mass.
10
temperature power spectrum [51, 52]. The temperature power spectrum on scales smaller than
the acoustic horizon at the recombination (l & 200) is relatively suppressed from the enhanced
Landau damping (not Silk damping) of the CMB photons. On the other hand, the polarization
power spectrum on scales larger than the acoustic horizon at the recombination (20 . l . 200)
is enhanced because of the increased probability of the Thomson scattering of CMB photons be-
tween the recombination and the reionization. The quantity constrained from CMB observations
is the energy injection rate per volume given as
dE
dtdV= feff ×∆E × n2
χ1〈σannvrel〉 , (11)
where ∆E ∼ mdm is the injected energy per annihilation, and feff is the efficiency of energy
deposition which is typically an order unity number depending on the annihilation product.
Assuming that the component annihilating into SM accounts for the total observed DM density,
the recent data from Planck [41] constrains DM annihilation as
feff〈σ1vrel〉rec
mχ1
. 3.2× 10−28 cm3 s−1 GeV−1 · (1/r1)2, (12)
where we scale the constraint with respect to r1, since both χ0 and χ1 contribute to the DM
density but only χ1 annihilates into SM particles. Hereafter, we take feff = 1. For simplicity, we
assume neither resonances nor non-perturbative enhancements of the annihilation cross section.
• DM annihilations in the Milky Way [53, 54]: DM annihilations in the Milky Way halo could
produce significant flux of diffuse X-ray and γ-ray photons. Therefore, the measured photon flux
from the satellite observations sets upper bounds on the annihilation cross section of DM. We
employ the bounds presented in Refs. [53, 54]; assuming that DM dominantly annihilates into
e+e−, the photon flux from final state radiation off the DM annihilations and the inverse Compton
scattering of the produced e+/e− with low energy photons (CMB, infrared light and starlight)
should be smaller than the observed one. In the case of s-wave annihilation of sub-GeV DM, the
CMB bounds on DM annihilation is roughly a few orders of magnitude stronger than the one
from the DM annihilations in the Milky Way halo. Since the upper bounds on DM annihilation
are basically given in term of the rate n2χ1
(σ1vrel)s at the Galactic velocity scales, we rescale the
bounds on DM annihilation cross section with respect to the factor r−21 .
We summarize the aforementioned indirect-detection constraints on DM annihilations in Figure 3
in the mχ1 versus r1 plane for a given mχ0 . At each point in the plane, we determine (σ1vrel)saccording to Eq. (5). The dotted curve in Figure 3 separates the two regimes, i.e., the standard and
assisted regimes, for the chemical freeze-out of χ1. As a reference, we plot the contours for the minimal
contribution to the elastic scattering cross section between χ1 and electron in the heavy mediator limit
σχ1e (dot-dashed) given by
σχ1sm ∼ (σ1vrel)s ×(µχ1sm
mχ1/2
)2
(13)
where µχ1sm is the reduced mass of the χ1-sm system. We also plot the direct-detection constraints
based on this minimal contribution (in brown):
• Direct-detection constraints on χ1-SM interaction: The elastic scattering cross section of
χ1 with SM can receive a minimal contribution given by Eq. (13); for concreteness, we assume
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
FIG. 3. Summary of constraints on s-wave annihilating χ1 in the absence of self-heating. In the assisted regime
(below the dotted curve), the annihilation cross section is sharply enhanced towards smaller r1, i.e., (σ1vrel)s ∝1/r2
1. Since the volumetric annihilation rate n2χ1
(σ1vrel)s is virtually independent of r1, the constraints on χ1-
annihilation is independent of r1 in the assisted regime. In the region that is not constrained by the MeV-scale
freeze-out [40] (green), i.e., mχ1 & 4.6 MeV, the strongest constraint are the bounds on DM annihilations
from observations on CMB [41] (sky-blue) and Galactic diffuse X-ray and γ-ray photons [53] (deep-blue); the
constraints rule out most of the parameter space of the sub-GeV two-component DM scenario. The constraints
from photo-dissociation of light nuclei does not appear in the presented parameter space. For reference, we also
display the contours for the minimal contribution to σχ1e in the heavy mediator limit [Eq. (13)] (dot-dashed),
and the corresponding direct-detection limits [55–58] (brown).
the heavy mediator limit. We present the direct-detection constraints on χ1-e scattering cross
section σχ1e in Figure 3 as a reference. We employ the direct-detection constraints on sub-
GeV DM from the following experiments (with the mass range where they are most sensitive):
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">AAADCXicjVHdStxAGD1GbdXaNuplb4JLYbcta1KKeiMI9sI7FdxVMEtIxtnsYP6YTARZ9gl8E+96V7z1BXpX9An0LfxmjFBXik5IcuZ855yZbyYqElEq172esCanpt+8nZmdezf//sNHe2GxW+aVZLzD8iSXh1FY8kRkvKOESvhhIXmYRgk/iE62dP3glMtS5Nm+Oit4Lw3jTPQFCxVRgf1TBp7/zU94XzU3/J2Ux2Ew9NlABN5opfmEcEdfxwQtX4p4oFqB3XDbrhnOc+DVoIF67Ob2X/g4Rg6GCik4MijCCUKU9BzBg4uCuB6GxElCwtQ5Rpgjb0UqToqQ2BP6xjQ7qtmM5jqzNG5GqyT0SnI6+EyenHSSsF7NMfXKJGv2f9lDk6n3dkb/qM5KiVUYEPuS71H5Wp/uRaGPddODoJ4Kw+juWJ1SmVPRO3f+6UpRQkGcxsdUl4SZcT6es2M8peldn21o6rdGqVk9Z7W2wp3eJV2wN36dz0H3e9tbbXt7PxqbX+qrnsEnLKNJ97mGTWxjFx3KvsAfXOPGOrd+Wb+tywepNVF7lvBkWFf3HRmqOQ==</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
FIG. 4. Same as Figure 3 but for p-wave annihilating χ1 while the hatched region at the right-bottom corner
is the unitarity bound on the DM annihilation cross section. In the assisted regime (below the dotted curve),
the annihilation cross section is sharply enhanced towards small r1, i.e., (σ1vrel)p ∝ 1/r31. The only robust
constraint in the presented parameters is the Neff constraint from MeV-scale freeze-out (green). Constraints
on (σ1vrel)p from photo-dissociation, CMB, and DM annihilations in the MW do not appear in the presented
range. For r1 < 0.01, the minimal s-wave contribution can become relevant. As an example, we present the
bounds coming from the unsuppressed s-wave contribution in the heavy mediator limit (the transparent regions
in sky-blue, deep-blue, and orange).
lighter χ0 due to the enhanced number density of χ0, and thus larger (σ1vrel)s is required to achieve the
desired r1. We see that for s-wave annihilating χ1, the strongest constraint on χ1-annihilation comes
from the CMB bound which disfavor the whole parameter space for the sub-GeV two-component DM
scenario.
2. p-wave annihilation of χ1
In the previous section, we have seen that the cosmological/astrophysical constraints disfavor s-wave
annihilating χ1 in the sub-GeV mass range, even for r1 1. If DM annihilation is p-wave suppressed,
the annihilation cross section may be small enough at the cosmological epochs of interest and therefore
sub-GeV DM can be consistent with the existing bounds. The difference from the s-wave annihilation
case is that in the assisted regime, the required annihilation cross section increases even more sharply
towards smaller r1, (σ1vrel)p ∝ 1/r31 [Eq. (10)]. The highly enhanced χ1-annihilation cross section for
r1 1 could render several caveats to be kept in mind on the χ1-SM interaction, as will be discussed
below. In the rest of this section, we will describe the thermal history of p-wave annihilating χ1, and
discuss the various constraints on χ1-annihilation described in the previous section.
For p-wave annihilating DM, the bounds on DM annihilation from the observations on light element
abundances and CMB depend on the DM temperature evolution during the relevant cosmological
13
epochs. We expand the annihilation cross section of χ1 in the non-relativistic limit as 5
〈σ1vrel〉 ' (σ1vrel)s + (σ1vrel)p 〈v2rel〉 . (14)
Note that what we mean by p-wave annihilation is that the term proportional to (σ1vrel)p is dominant
around T = Tfo,1 ∼ mχ1/20. Even in the p-wave annihilation case, the unsuppressed s-wave annihila-
tion contribution (σ1vrel)s may be dominant over the other around the cosmological epoch of interest
for r1 1, as will be discussed shortly. In the absence of the DM self-heating epoch, the temperature
evolution of χ1 is
Tχ1 =
T forT > Tkd ,
Tkd [a(Tkd)/a(T )]2
forT < Tkd ,(15)
where Tkd is the SM temperature at the kinetic decoupling of χ1 and a is the scale factor. Hereafter,
we assume that the elastic scattering process that keeps χ1 in kinetic equilibrium, i.e., χ1sm→ χ1sm,
is related to the p-wave annihilating process of χ1 by the crossing symmetry. In the heavy mediator
limit, the elastic scattering cross section has the minimal contribution given by Eq. (13).
If the kinetic decoupling of χ1 takes place before the electron-position annihilation, T & me/20,
the decoupling point is in turn virtually determined by the χ1e → χ1e process. For r1 1, due to
the enhanced annihilation cross section (and thus the enhanced σχ1sm), the kinetic decoupling could
happen after the electron-position annihilation. In such a case, the elastic scattering of χ1 with proton
also has to be taken into account. We determine the kinetic-decoupling temperature Tkd is determined
by the condition γχ1sm ' H where γχ1sm is the momentum transfer rate given by [63–65]
γχ1sm '(δE
T
)nsmσχ1sm 〈vrel,χ1sm〉 , (16)
where δE is the change in χ1 kinetic energy per elastic scattering and 〈vrel,χ1sm〉 is the averaged
relative scattering velocity between χ1 and an SM particle. For elastic scattering with electrons, we
may estimate δE/T as ' T/mχ1 (' me/mχ1) for relativistic (non-relativistic) electrons. For the
scattering with non-relativistic protons, δE/T ' mχ1/mp. For general Tχ1
, the relative scattering
velocity is given as
〈vrel,χ1sm〉2 =8
π
(Tχ1
mχ1
+T
msm
), (17)
where we may put Tχ1 = T when estimating Tkd in the absence of DM self-heating; if the χ1 exhibits
the self-heating epoch, the kinetic-decoupling point can be determined by a different condition, as will
be discussed in the next section.
Since the photo-dissociation constraints are sensitive to the DM annihilation rate n2χ1〈σ1vrel〉, the
constraints depends on the DM temperature evolution in the temperature range relevant to photo-
dissociation of light nuclei, 100 eV . T . 10 keV. Therefore, in order to put the photo-dissociation
constraints on χ1-annihilation, one needs to estimate Tkd. For Tkd & 10 keV, the redshift behavior
is Tχ1 ∝ 1/a2 during the relevant epoch and we simply rescale the photo-dissociation constraints (as
an upper bound) on (σ1vrel)p for Tkd = 10 keV [36] with the factor ∼ (Tkd/10 keV) (aside from the
5 Note that 〈σ1vrel〉 represents the total annihilation cross section and does not specify a final state. Dominant annihi-
lation processes for (σ1vrel)s and (σ1vrel)p may have different final states. For p-wave annihilating χ1, the dominant
annihilation channels for the s and p-wave contributions may be given as in Figure 5.
14
χ1
χ1
sm
sm
χ1smsm
χ1smsm
FIG. 5. 2-body (left) and 4-body (right) annihilation channels of χ1. While the p-wave 2-body annihilation
channel of χ1 dominates the annihilation of χ1 around the freeze-out of χ1, the unsuppressed s-wave 4-body
annihilation channel may become relevant afterwards, e.g., during the photo-dissociation epoch, recombination
epoch, and inside the MW halo.
rescaling with r1 discussed above). For Tkd . 100 eV, since the redshift behavior is Tχ1 = T during
the relevant epoch, we may simply take the upper bound on (σ1vrel)p for Tkd = 100 eV. If Tkd lies
within the range 100 eV . T . 10 keV, we aggressively underestimate the upper bound by applying
the same upper bound with the Tkd = 100 eV case (orange region with dashed boundary in Figure 4);
this is to display the potentially constrained parameter region, while a robust bound requires dedicated
analyses.
The CMB bounds on DM annihilations are also sensitive to the DM annihilation rate at the last
scattering and one needs to evaluate the DM temperature around the recombination epoch T ∼0.235 eV. For the CMB bounds on χ1-annihilation, we estimate Tχ1 at the CMB epoch, T = 0.235 eV
using Eq. (15). The Galactic χ1-annihilations could provide stronger upper bounds on (σ1vrel)p than
the CMB bound since the annihilation rates can be larger in the Galactic halo compared to the one
in the recombination epoch; this is because the velocities of DM particles in the Galactic halo can be
larger than the DM velocities around the recombination. We take 〈vrel〉 ∼ 220 km/s to estimate the
annihilation cross section on the Galactic scales [53].
As we have done in the case of s-wave annihilating χ1, we summarize the aforementioned indirect-
detection constraints in Figure 4. As a reference, we plot the contours for the minimal contribution
to σχ1e (dot-dashed) according to Eq. (13) [but with (σ1vrel)p instead of (σ1vrel)s] and present the
corresponding direct-detection constraints based on this minimal contribution (brown). In the assisted
regime, the annihilation cross section is enhanced for small r1 as (σ1vrel)p ∝ 1/r31. Since the volumetric
annihilation rate scales as n2χ1〈σ1vrel〉 ∝ 1/r1 for p-wave annihilation, the constraints are more relevant
towards the small r1.
We find that for p-wave annihilating χ1, the only robust constraint appearing in Figure 4 is the Neff
bound from the MeV-scale freeze-out of DM (in green). However, for r1 < 0.01, the unsuppressed
s-wave component (σ1vrel)s can be dominant over the p-wave part during the cosmological epoch of
interest. In the heavy mediator limit, we may have the following minimal contribution to (σ1vrel)sgiven by
(σ1vrel)s ∼m2χ1
(4π)3(σ1vrel)
2p , (18)
where we have in mind the unsuppressed 4-body annihilation channel contributing to (σ1vrel)s (see
Figure 5). We plot the possible constraints from the minimal s-wave contribution as well (labeled by
15
‘minimal s-wave’). The photo-dissociation constraint with the dashed boundary is the region where
the s-wave contribution starts to dominate during the relevant photo-dissociation epoch, 100 eV .
T . 10 keV; in such a case, we take the more constraining bound among the pure s-wave case and the
pure p-wave case.
III. SELF-HEATING FROM BOOSTED DM PARTICLES
After the chemical freeze-out of χ0, residual annihilations of χ0 continuously produce boosted χ1
particles. Before the kinetic decoupling of χ1, the boosted χ1 particles have no effect on the evolution
of Tχ1 . As we have discussed in the previous section, the mere effect of the produced χ1 is to contribute
to the relic abundance of χ1. However, if χ1 exhibits sizable self-scattering so that the self-scattering
is efficient even after the kinetic decoupling of χ1, the temperature evolution of χ1 after the kinetic
decoupling exhibits interesting dynamics. In the presence of efficient self-scattering, the excess kinetic
energy of energetic χ1 particles produced from residual χ0-annihilations are shared with the majority
of the χ1 particles and heat the χ1 particles as a whole. Such processes, which we dub as the DM self-
heating, could enhance the temperature of χ1 compared to the SM one. For example, if χ1 elastically
scatter with electrons, the kinetic decoupling typically occurs around the electron-position annihilation
due to dwindling electron number density. 6 Assuming Tχ1= T , the decoupling of self-scattering takes
place when the SM temperature is
Tdec,self 'me
20
( mχ1
100 MeV
)1/3(
0.1
r1
)2/3(10−6 cm2/g
σself/m
)2/3
, (19)
where σself/m is the self-scattering cross section per χ1 mass and me/20 is the SM temperature around
the electron-positron annihilation. Thus, if the self-scattering cross section is large enough to delay
the decoupling of self-scattering beyond the kinetic-decoupling point, DM undergoes self-heating until
the decoupling of self-scattering. After then, χ1 particles adiabatically cool as Tχ1 ∝ 1/a2.
In this section, we demonstrate the cosmological evolution of χ1 in the case of s-wave annihilation of
χ0 and p-wave annihilation of χ1; as we have discussed in the last section, the case of s-wave annihilation
of χ1 is strongly disfavored by the CMB bounds on DM annihilation. With the s-wave annihilation
of χ0, the temperature of χ1 could redshift like radiation Tχ1∝ 1/a even after the kinetic decoupling.
The modified evolution of Tχ1 from the self-heating adds several interesting aspects to the cosmological
constraints on χ1. For self-scattering cross section of χ1 as large as σself/m ∼ 0.1 cm2/g, the self-heating
epoch could persist until the matter-radiation equality. Such elongated self-heating epoch may suppress
the structure formation of χ1 and could be subject to the warm dark matter (WDM) constraints, e.g.,
from the Lyman-α forest observations. The warmness of χ1 could also suppress the clustering of χ1 in
the Galactic scale and may relax the direct/indirect-detection constraints. Enhanced Tχ1 also affects
the constraints that directly depend on the annihilation rate of χ1. For example, if the self-heating
epoch overlaps with the epoch relevant for photo-dissociation of light nuclei, the constraints on the
annihilation cross section would become severer.
We describe the self-heating of χ1 in Section III A; details of the Boltzmann equations and the
analytic arguments are collected in the Appendix A. We discuss the implications of the DM self-
heating epoch on the cosmological constraints in Section III B.
6 A notable exception is when the final DM abundance is set by the DM annihilation through a resonant mediator [66];
while the annihilation cross section is resonantly enhanced, the DM-SM elastic scattering is relatively suppressed and
the kinetic decoupling may take place very close to the freeze-out.
16
A. Thermal history of χ1 with self-heating
Efficient self-scattering of χ1, i.e., Γself & H, allows χ1 particles to efficiently exchange their energy
and momentum among themselves. Regardless of the energy exchanges with external systems, efficient
self-scattering forces χ1 particles to follow the thermal energy distribution fχ1(E) ∝ e−E/Tχ1 ; this is
the case even in the presence of the χ0χ0 → χ1χ1 process. After the chemical decoupling of χ0,
residual χ0-annihilations produce a minority of boosted χ1 particles. Efficient self-scattering quickly
redistribute the excess kinetic energy to the majority of χ1 particles, heating the χ1 particles as a
whole. In such a case, the evolution of χ1 temperature is described by the following equation:
Tχ1+ 2HTχ1
' γheatT − 2γχ1sm (Tχ1− T ) , (20)
where γheat is defined as
γheat =2n2
χ0(σ0vrel) δm
3nχ1T. (21)
After the chemical freeze-out, the abundances of χ0 and χ1 are virtually conserved and thus Eq. (20)
alone determines the evolution of Tχ1. Note that we have assumed that both χ0 and χ1 are non-
relativistic in Eq. (20) (see Appendix A for details). The inverse of the heating rate, γ−1heat, represents
the timescale during which a χ1 particle obtains kinetic energy comparable to ∼ T . The two terms
in the RHS of Eq. (20) represents the two paths for the energy exchange of χ1 with external systems.
The term proportional to γχ1sm represents the energy exchange with the SM plasma through the
χ1sm→ χ1sm process.
Initially, γχ1sm is dominant over both H and γheat, and the kinetic equilibrium is achieved. As the
Universe cools, γχ1sm drops and the term proportional to γχ1sm could become negligible from Eq. (20).
The self-heating epoch starts from then, and the heat injection from χ0-annihilation can modify the
evolution of Tχ1from what we expect for free-streaming non-relativistic particles, i.e., Tχ1
∝ 1/a2. In
the case of s-wave annihilation of χ0, the temperature ratio Tχ1/T asymptotes to the following:(
Tχ1
T
)asy.
∼ γheat
H,
'
2 (1− r1)
3r1
mχ1δm
mχ0Tfo,0
(g? (Tfo,0)
g? (Tasy.)
)1/2g?S (Tasy.)
g?S (Tfo,0)for T > Teq ,
4 (1− r1)
3r1
mχ1δm
mχ0Tfo,0
(g? (Tfo,0)
g? (Teq)
)1/2g?S (T )
g?S (Tfo,0)
(T
Teq
)1/2
for T < Teq ,
(22)
where we have used Eq. (2) and Teq ∼ 0.75 eV is the SM temperature at the matter-radiation equality.
In the case of p-wave annihilation of χ0, Tχ1scales as the Tχ1
∝ 1/a2, while there is an enhancement
compared to the case of no self-heating (see Appendix C for the discussion on the case of p-wave
annihilation of χ0). Hereafter, we focus on the case of s-wave annihilation since it exhibits the maximal
impact of self-heating.
Due to practical reasons, we do not attempt to follow the full evolution of Tχ1 from Eq. (20). Instead,
we specify an interval in T where we can reliably estimate Tχ1:
The first term in the RHS is the friction term for Tχ1and the second term is the source term. γheat
is multiplied by the unit-step function by hand to incorporate the stopping of boosted χ1 by the SM
plasma; if the boosted χ1 particles dominantly scatter with SM particles, self-heating is ineffective. We
remark that such SM-stopping of χ1 is only relevant for small r1 0.1, where large annihilation cross
section is required deplete χ1 to the desired r1. Since we consider σself/m as strong as ∼ 0.1 cm2/g,
boosted χ1 dominantly scatters with χ1 for r1 & 0.1. When evaluating the rates for boosted χ1 in the
step-function, we set vrel = c for simplicity.
The asymptotic solution given by Eq. (22) is defined when γχ1sm is negligible from Eq. (24). We
define Tdec,el as the SM temperature below which γχ1sm is negligible as a friction term, i.e, γχ1sm . H.
We define Tsh as the SM temperature below which γχ1sm becomes negligible as a source term; according
to the definition, we have
Tsh = min [Tstop, Theat] , (25)
where Tstop and Theat are determined by the condition Γself,vrel=c = Γχ1sm,vrel=c and γheat = γχ1sm,
respectively. Therefore, the asymptotic solution for Tχ1is defined for T . Tmin, where Tmin is given
18
by
Tmin = min [Tdec,el, Tstop, Theat] . (26)
Note that to determine the true value of Tmin, one needs a priori knowledge on the exact evolution of
Tχ1, since the rate γχ1sm generally depends on Tχ1
. Nevertheless, we underestimate Tmin as follows;
we overestimate γχ1sm by assuming the maximal temperature of Tχ1= (Tχ1
/T )asy.T to underestimate
Tdec,el and Theat. At temperatures lower than the underestimated Tmin, one can reliably estimate Tχ1
with Eq. (23), regardless of the exact evolution of Tχ1for T & Tmin. By taking into account the
estimation given in Eq. (23), we modify the estimation of Tdec,self from Eq. (19) as
Tdec,self ' 1 eV
(Tχ1
T
)−nasy.
(0.3
r1
)2n ( mχ1
100 MeV
)n( 1 cm2/g
σself/mχ1
)2n
, (27)
where n = 1/3 for when Tdec,self > Teq, and n = 2/9 for when Tdec,self < Teq. If Tdec,self evaluated
in this way is larger than Tmin, our estimation of Eq. (23) is not self-consistent and thus not reliable.
Hereafter, we use Eq. (23) to estimate Tχ1 when the consistency condition Tmin > Tdec,self is satisfied
at the most conservative level; we underestimate Tmin by taking the highest possible value for Tχ1,
i.e., the asymptotic solution Eq. (22). At the same time, we overestimate Tdec,self by taking the lowest
possible value, i.e., the evolution in the absence of DM self-heating Eq. (15) (see Appendix D for more
discussion). In Figure 6, we present the numerical solutions to Eq. (24); for temperatures lower than
the underestimated Tmin (green circles), we find that Eq. (23) approximates well the evolution of Tχ1 .
On the other hand, there may be cases where γheat never becomes dominant in the source term before
the decoupling of self-scattering, i.e., Tsh < Tdec,self . In such a case, Tχ1 is reliably estimated with
Eq. (15). Again, since we cannot a priori determine Tsh and Tdec,self before knowing the evolution of
Tχ1 , we conservatively overestimate (underestimate) Tsh (Tdec,self). More specifically, we determine Tsh
(Tdec,self) by taking the lowest (highest) possible values for Tχ1, which is estimated by Eq. (15) [Tχ1
=
(Tχ1/T )asy.T ].
B. Cosmological constraints on DM self-heating
1. Warm dark matter constraints
Before the matter-radiation equality, Tχ1redshifts like radiation during DM self-heating epoch. For
σself/m as large as ∼ 1 cm2/g, DM self-heating epoch could persist until the vicinity of the matter-
radiation equality, i.e., Tdec,self ∼ Teq [Eq. (27)]. The resultant Tχ1 around Teq is much larger than
that without DM self-heating, and may be sizable to make χ1 behave as warm dark matter (WDM).
Therefore, in the presence of DM self-heating, the total relic dark matter is composed of two com-
ponents with distinct temperatures: warm χ1 and cold χ0. One way to represent the warmness of
DM is the cutoff in the resultant matter power spectrum, which can be estimated by the (co-moving)
Jeans scale kJ at the matter-radiation equality. kJ is the wave number that appears in the evolution
equation of χ1’s density perturbation, and corresponds to the length scale λJ = 2π/kJ below which
the pressure gradient of DM wins over gravity. For density perturbations of wave numbers k > kJ,
χ1 cannot experience gravitational collapse due to its own velocity dispersion. The reason that kJ is
evaluated at the matter-radiation equality is that DM density perturbations start to rapidly grow only
after the matter-radiation equality, and kJ of χ1 takes the minimum value (the largest length scale)
19
then since kJ ∝ a1/2 during the matter-dominated era. Assuming the temperature evolution of χ1
follows the estimation given in Eq. (23) 7, the Jeans wave number of χ1 is given as
kJ = a
√4πGρm〈−→v 2〉1
∣∣∣∣eq
,
' 76 Mpc−1
(r1
1− r1
)1/2 (mχ0
δm
)1/2 ( mχ0
100 MeV
)1/2
max
[1,
√Tdec,self
Teq
],
(28)
where ρm is the average matter density, and 〈−→v 2〉1 is the variance of χ1 velocity. We remark that
although the kJ of χ1 implicitly depends on mχ1, the value of mχ1
itself is not important; this is because
〈−→v 2〉1 ∝ Tχ1/mχ1 = (Tχ1/T )asy.(T/mχ1) and the additional factor of mχ1 from the (Tχ1/T )asy.
through DM self-heating [Eq. (22)] cancels the explicit mχ1dependence. For a fixed δm, kJ of χ1
increases towards heavier χ0; this is because heavier χ0 corresponds to smaller χ0 number density and
hence smaller heating rate [Eq. (21)].
On the other hand, kJ alone cannot entirely represent the overall effect of DM self-heating on
structure formation. This is because we have an additional parameter, r1; no matter how kJ is small,
warmness of χ1 would have negligible effect on the overall matter power spectrum for r1 1. Therefore
the two parameters, kJ and r1, are needed to characterize the resultant matter power spectrum. 8 Given
the abundance ratio r1, we investigate how kJ is constrained by observations.
As shown in Eq. (28), the cutoff scale defined by the linear matter power spectrum could be at the
galactic scales, i.e., kJ = O (1) Mpc−1, and thus χ1 could behave as WDM. To differentiate WDM
from cold dark matter (CDM), it is better to look into the matter distribution at high redshifts or
the abundance of sub-galactic scale non-linear objects. This is because the formation of large-size
halos enhance the small-size correlation in the non-linear matter power spectrum to compensate the
original discrepancy of WDM from CDM (in the linear matter power spectrum). Also, the abundance
of small-size gravitationally bound objects is sensitive to the linear matter power spectrum before the
non-linear growth of structures [69]. We summarize the considered WDM constraints below; we choose
these observations since the constraints are explicitly given in terms of the mixed DM scenarios.
• Lyman-α forest observations [70] : One of the most stringent constraint on the warmness of
DM comes from the observations on rather high redshifts, i.e., z ∼ 3. As discussed above, it is
more advantageous to look into the structure of the Universe at higher redshifts to discriminate
WDM and CDM. One of the promising methods is the Lyman-α forest method. Lyman-α
absorption lines in the spectrum of distant quasars can be used as a tracer of cosmological
fluctuations on scales k ∼ 0.1–10hMpc−1, at redshifts z = 2–4. We translate the constraints
for warm+cold DM (or mixed DM) into our scenario. For example, in Fig. 6 of Ref. [70], the
constraints on mixed DM is given in the mwdm versus rwarm plane, where mwdm is the mass
of conventional thermal WDM and rwarm is the fraction of them in mass density. What we
mean by conventional thermal warm DM is that the DM particles of mass mwdm,th follow the
7 We note that the WDM constraints we will consider will only be relevant for r1 & 0.07; for such r1, estimation of Tχ1
given in Eq. (23) is a good approximation and the WDM constraints with respect to Eq. (28) will be robust.8 We remark that one may use a different definition of kJ to estimate the suppression scale in mixed warm+cold DM
scenarios. Instead of taking the velocity dispersion of χ1 in Eq. (28), we may use the velocity dispersion of total DM,
i.e., 〈−→v 2〉tot. ' r1〈−→v 2〉1 [67]. The Jeans wave number defined with the r1-weighted velocity dispersion represents
the scale at which the matter power spectrum exhibits sizable suppression from the CDM one, while the one defined
in Eq. (28) represents the exact suppression scale from the CDM one. Nevertheless, once we specify r1, the WDM
constraints are the same as long as we consistently use one definition of kJ for mixed DM scenarios. See also Ref. [68]
for discussion on the Jeans scale of DM in the case of multiple species with distinctive distribution function behaving
as WDM.
20
Fermi-Dirac distribution with temperature Twdm,th (motivated by, e.g., light gravitino DM from
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">AAADCXicjVHdStxAGD1GbdXaNuplb4JLYbcta1KKeiMI9sI7FdxVMEtIxtnsYP6YTARZ9gl8E+96V7z1BXpX9An0LfxmjFBXik5IcuZ855yZbyYqElEq172esCanpt+8nZmdezf//sNHe2GxW+aVZLzD8iSXh1FY8kRkvKOESvhhIXmYRgk/iE62dP3glMtS5Nm+Oit4Lw3jTPQFCxVRgf1TBp7/zU94XzU3/J2Ux2Ew9NlABN5opfmEcEdfxwQtX4p4oFqB3XDbrhnOc+DVoIF67Ob2X/g4Rg6GCik4MijCCUKU9BzBg4uCuB6GxElCwtQ5Rpgjb0UqToqQ2BP6xjQ7qtmM5jqzNG5GqyT0SnI6+EyenHSSsF7NMfXKJGv2f9lDk6n3dkb/qM5KiVUYEPuS71H5Wp/uRaGPddODoJ4Kw+juWJ1SmVPRO3f+6UpRQkGcxsdUl4SZcT6es2M8peldn21o6rdGqVk9Z7W2wp3eJV2wN36dz0H3e9tbbXt7PxqbX+qrnsEnLKNJ97mGTWxjFx3KvsAfXOPGOrd+Wb+tywepNVF7lvBkWFf3HRmqOQ==</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=
1/(
0+
1))
FIG. 8. Same as Figure 4 but in the presence of self-heating of χ1. Due to the self-heating of χ1, the WDM
constraints on χ1 emerges [see Figure 7] (pink). Furthermore, the constraints on χ1-annihilation from photo-
dissociation, and CMB are modified from Figure 4; we employ Eq. (23) and Eq. (15) to estimate Tχ1 during
the relevant cosmological epochs. The constrained region with solid boundaries are the ones where we can
reliably estimate Tχ1 in the relevant cosmological epochs through Eq. (23) and Eq. (15). The constrained
region with dashed boundaries are the ones where Eq. (23) is not a robust estimation for Tχ1 . On the other
hand, for r1 < 0.01, the minimal s-wave contribution can become relevant. We over plot the CMB bound
coming from the unsuppressed s-wave contribution in the heavy mediator limit [see Figure 4] (gray). We also
plot the direct-detection constraints on the minimal contribution to σχ1e in the heavy mediator limit (brown);
the region enclosed by the dashed brown curve is also the region constrained by direct-detection experiments,
but χ1 may not cluster on the Galactic scales. The evolution of Tχ1 for the parameters depicted by stars are
presented in Figure 6.
25
IV. IMPACT ON DARK PHOTON SEARCHES
The light DM component χ1 can be directly probed in high-intensity accelerator experiments [79–82],
which provides a complementary approach in identifying the multi-component dark matter scenarios.
In order to show such a complementarity, we fix a reference model where both χ0 and χ1 are the SM
gauge singlet complex scalars and a dark photon A′ mediates the interaction between χ1 and the SM
sector. The relevant terms in the effective Lagrangian are:
L ⊃ εA′µJµem − igDA′µ (χ∗1∂µχ1 − χ1∂
µχ∗1)− λast.
4|χ1|2 |χ0|2 , (35)
where mA′ is the dark photon mass and gD is the dark gauge coupling. The dark photon A′ kinetically
mixes with the SM photon and induces the coupling with the SM electromagnetic current Jµem, which
is set by the kinetic mixing parameter ε. As will be discussed in more detail, we found the parameter
region of mA′−ε which is expected to be reached by the current and future experiments can be sensitive
to our WDM constraints for 0 . r1 . 1.
There are several tree-level annihilation channels for χ1 that may determine χ1 relic abundance, e.g.,
χ1χ∗1 → ff and χ1χ
∗1 → A′A′. The former annihilation channel is generated through the kinetic mixing
of A′; the χ1 pair annihilates through an off-shell A′ and it is p-wave suppressed for a complex scalar
χ1. This is the dominant number changing process for the case of mA′/mχ1> 2, while χ1χ
∗1 → A′A′
is kinematically forbidden. We focus on such a case throughout this section. For 1 < mA′/mχ1 < 2,
if χ1χ∗1 → ff determines the relic density of χ1, while the unsuppressed s-wave annihilation χ1χ
∗1 →
A′ff around the last scattering strongly disfavor this case. 11 The relic abundance of χ0 particles
is determined through the λast. coupling, i.e., the χ0χ∗0 → χ1χ
∗1 process which is s-wave. The non-
relativistic annihilation cross sections of χ1 and χ0 are given as(σχ0χ∗0→χ1χ∗1vrel
)' λ2
ast.
32πm2χ0
√1−R−2
χ0 , (36)
(σχ1χ∗1→ffvrel
)' 4πααDε
2
3
∑f
q2f
(2 +R2
f
)√1−R2
f
m2χ1
(R2A′ − 4)
2 ×(s/m2
χ1− 4), (37)
where qf is the EM charge of the SM fermions, αD = g2D/4π, and Ri = mi/mχ1
. The factor of
(s/m2χ1− 4) in Eq. (37) may be replaced with 〈v2
rel〉 ' 6Tχ1/mχ1
when we take the thermal average.
We present the thermal relic curves for given values of the mass ratio mA′/mχ1 (black) in Figure 9
for various values of r1 in the mA′ versus ε2 plane. We also over plot the constraints on ε from the
low-energy experiments (gray), which search for missing-energy/momentum events via the production
of dark photons. One of the relevant constraint comes from the NA64 collaboration [84], which is
the missing-energy experiment. It is based on the detection of the missing energy carried away by
the soft production of A′ by scattering high-energy electrons to the active beam dump target (via
bremsstrahlung emission of A′ in the process e−Z → e−ZA′); as we focus on mA′/mχ1> 2, dark
photon decays invisibly, i.e., A′ → χ1χ∗1, with the branching ratio close to unity. The difference of
this type of experiment from the conventional beam-dump experiments 12 (see, e.g., Ref. [85, 86]),
11 The dark sector processes like the χ1χ1χ∗1 → χ1A′ and χ1χ∗1 → A′A′ could also determine the relic density of χ1 [83],
in which case the stringent CMB bounds may be evaded.12 There, A′ is produced by a high-intensity beam in a dump and generate a flux of DM particles through the A′ → χχ
decay. The produced DM through the decay could be detected through the scattering off electrons in the far target.
26
is that there is no need for additional DM scattering at a far target. Therefore, the sensitivity is
proportional to the production cross section of A′, which scales as ∝ ε2/m2A′ for a given mass ratio
mA′/mχ1 ; this is why the thermal relic curves in the standard freeze-out regime (top and center panels
of Figure 9) are nearly parallel to the lower boundary of the constraint from NA64. In the bottom
panel, the break of the thermal relic curves represents the transition to the assisted freeze-out regime,
since the required annihilation cross section also depends on mχ1in the assisted regime [Eq. (10)]. We
also display the constraint from the BaBar collaboration [87], which searches for events with a single
high-energy photon and a large missing momentum and energy that is consistent with hard production
of A′ through the process e−e+ → γA′ followed by A′ → χ1χ∗1. The production cross section of A′ is
proportional to ∝ ε2/s and thus the sensitivity is virtually independent of mA′ .
We remark that, as can be seen in the bottom panel of Figure 9, the constraint from NA64 disfavors
the abundance ratio smaller than r1 . 0.1 unless the annihilation process χ1χ∗1 → ff is near the
resonance to push the required ε to smaller values, i.e., mA′/mχ1 → 2; investigating the robust
thermal relic curve near the resonance may require dedicated analyses [88]. Therefore, we focus on
r1 & 0.1 where the only relevant cosmological constraint from DM self-heating is the WDM constraint
on χ1.
The χ1 particles exhibit self-scattering via the A′-exchange. The self-scattering cross section of χ1
is given by
σself/mχ1 =6πα2
Dmχ1
m4A′
+3λ2
χ1
16πm3χ1
, (38)
where the second term in the RHS is the possible contribution to χ1 self-scattering from the λχ1|χ1|4
coupling. Hereafter, we set λχ1 = 0 for the simplicity of the discussion. 13 After the freeze-out of
DM, the residual annihilation of χ0 produces boosted χ1 particles and induces DM self-heating in
collaboration with the χ1 self-scattering. For r1 close to unity (the top panel of Figure 9), the χ0-
annihilation rate is suppressed. Thus the effect of DM self-heating is not significant [Eq. (22)] and the
WDM constraints vanish (also see Figure 7). Meanwhile, the self-scattering among χ1 particles can
be as large as σself/m ∼ 1 cm2/g; see the contours for σself/m in Figure 9 (dotted). Since χ1 is the
dominant component of DM, the large self-scattering among χ1 may conflict with the observations on
galaxy clusters [89–91]. As a reference, we display the constraint on σself/m from the Bullet cluster
based on mass loss (blue) [89] which covers the region not yet constrained by NA64.
As we consider smaller r1, only a sub-dominant component of DM exhibits self-scattering and the
constraint from the Bullet cluster may get relaxed. At the same time, the effect of DM self-heating
becomes more relevant. In the center and bottom panel of Figure 9, the WDM constraints of χ1 emerge
(pink), redeeming the relaxed constraint on the self-scattering cross section. The WDM constraints also
depend on mχ0 since the annihilation rate decreases as we consider larger mχ0 ; the WDM constraints
virtually vanish for mχ0& 300 MeV. As we further decrease r1 . 0.1, the WDM constraints are
virtually vanishing while the constraint from NA64 disfavors smaller values of r1.
13 For λχ1 = O(1), σself/mχ1 will considerably increase and hence lead to stronger WDM constraints (pink) in Figure 9.
Nevertheless, the qualitative discussions do not change.
27
BaBar
10cm
2/g
1cm
2/g
0.1cm
2/g
BaBar
BaBar
10cm
2/g
1cm
2/g
10cm
2/g
1cm
2/g
mA0/m
1=
2.5mA
0/m1
=5
mA0/m
1=
2.5
mA0/m
1=
5mA0/m
1=
10
mA0/m
1=
2.5
mA0/m
1=
5mA0/m
1=
10
50
50
NA64
NA64
NA64
0.1cm
2/g
10
BaBar
10cm
2/g
1cm
2/g
0.1cm
2/g
BaBar
BaBar
10cm
2/g
1cm
2/g
0.1cm
2/g
10cm
2/g
1cm
2/g
0.1cm
2/g
mA0/m
1=
2.5mA
0/m1
=5
mA0/m
1=
2.5mA
0/m1
=5mA
0/m1
=10
mA0/m
1=
2.5
mA0/m
1=
5mA0/m
1=
10
50
50
NA64
NA64
NA64
10
FIG. 9. Collection of various constraints on the toy model with dark photon portal [Eq. (35)] for various r1.
Along the black curves, desired thermal relic abundance of χ1 is achieved; in the top and the center panel,
the presented curves are in the standard freeze-out regime; in the bottom panel, the break of the thermal
relic curves represent the transition to the assisted freeze-out regime. The gray regions are constrained by
the missing energy/momentum experiments [84, 87]. The green region represents the Neff constraint from the
MeV-scale freeze-out of χ1 [40]. (Top): χ1 is the dominant component of DM and thus the WDM constraint is
vanishing. Meanwhile, the large self-scattering among χ1 provides a constraint [89] (blue) complementary to the
missing energy/momentum experiments. (Center): As we consider smaller r1, the self-scattering constraint
becomes relaxed, while the effect of DM self-heating become more relevant. The WDM constraint on χ1
emerges [70, 76] (pink). (Bottom): Larger annihilation cross section is required as we consider smaller r1 and
eventually conflicts with the constraints from NA64 for r1 . 0.1.
28
V. CONCLUSIONS
We have studied the cosmology of the two-component DM scenario, which serves as an illustrating
example of a non-minimal dark sector. In this scenario, two stable components, i.e., χ0 and χ1, consist
DM and only the lighter state χ1 interacts with SM. If χ1 interacts sufficiently strong with SM, the
dark sector may be in thermal equilibrium in the early Universe, and the DM relic density would be
determined through the thermal freeze-out of DM, i.e., through the χ0χ0 → χ1χ1 and the χ1χ1 →sm sm processes. We have carefully studied the dynamics of the two-component DM scenario, especially
focusing on the detectability of the sub-dominant component of DM, χ1, in cosmological/astrophysical
observations.
We have shown that as we consider a smaller χ1 abundance fraction, i.e., r1 . 0.1, the freeze-out of χ1
transits to the assisted-regime where the required annihilation cross section of χ1 is sharply enhanced
towards smaller values of r1. Contrary to the usual case of the standard thermal freeze-out of DM
where the annihilation cross section scales as ∝ 1/r1, the annihilation cross section of χ1 in the assisted-
regime scales as (σ1vrel)s ∝ 1/r21 in the case of s-wave annihilation, and as (σ1vrel)p ∝ 1/r3
1 in the case
of p-wave annihilation. The sharp scaling of the annihilation cross section implies better detectability
of χ1 for smaller abundance fractions, e.g., in direct/indirect-detection experiments. Having in mind
the sharp scaling, we have reviewed the cosmological/astrophysical constraints on χ1-annihilation.
It is worthwhile to note that considering smaller values of r1 in the two-component DM scenario is
sometimes considered to be a minimal remedy to evade the stringent constraints on sub-GeV DM
annihilations; however, considering smaller values of r1 does not relax the constraints.
We have demonstrated that self-scattering among χ1 could considerably affect the detectability of
χ1. The collaboration of the residual χ0-annihilation and the self-scattering leads to DM self-heating,
which may enhance the temperature of χ1 compared to the SM one. The self-scattering cross section
as large as σself/m ∼ 0.1 cm2/g can be naturally realized for a sub-GeV mass scale, and we have
shown that WDM constraints from the Lyman-α forest data and the number of satellite galaxies
in the MW are significant for mχ0. 200 MeV and r1 & 0.1. For abundance fraction smaller than
r1 . 0.1, although the warmness (or Jeans mass) of χ1 increases towards smaller r1, such a sub-
dominant fraction of χ1 have a negligible effect on the overall matter power spectrum, and hence
the warmness is not constrained by the cosmological observations on the structure formation of our
Universe. Nevertheless, the warmness of sub-dominant fraction of χ1 has interesting implications on
the interpretation of direct-detection experiments. For r1 0.1, the elastic scattering rate of an SM
particle with χ1 increases towards smaller r1 and thus direct-detection constraints on χ1 is expected to
be severer at the first sight. However, we have shown that the resultant warmness of χ1 could suppress
χ1’s gravitational clustering in our Galaxy and thus relax the direct-detection constraints. How much
the direct-detection constraints are relaxed depends on the suppression of the Galactic abundance
fraction of χ1 compared to the cosmological one. We have aggressively estimated the Galactic χ1
abundance fraction to vanish when the Jeans mass of χ1 exceeds the mass of the MW. However, we
expect the suppression of the Galactic abundance with respect to an increasing Jeans mass to be more
gradual. It would be interesting to investigate the gravitational clustering of χ1 at the non-linear level
and put more robust direct-detection constraints.
Moreover, DM self-heating may enhance χ1 temperature during cosmological epochs sensitive to
DM annihilations, e.g., during the photo-dissociation epoch of light nuclei (100 eV . T . 10 keV) and
at the last scattering. In the case of p-wave annihilation of χ1, the enhanced χ1 temperature from DM
self-heating increases the annihilation rate, and we have demonstrated that the cosmological constraints
29
on χ1-annihilation could become relevant for r1 0.1. We have aggressively displayed the parameter
regions that can be potentially constrained by the BBN and CMB observations. Interestingly, such
parameter regions redeem the relaxed direct-detection constraints for r1 0.1. Therefore, it would
be interesting to do more robust analyses of the cosmological constraints of DM annihilations.
Despite the interesting cosmology for r1 0.1, we remark that for such small values of r1, the
χ1-SM interaction is highly enhanced and it is usually incompatible with accelerator-based experi-
ments. We have demonstrated this by taking a case where χ1 interacts with SM through the dark
photon portal. We have focused on the case of mA′/mχ1> 2, where the missing-energy/momentum
experiments provide relevant constraints on the kinetic mixing parameter. In particular, the missing-
energy experiment at NA64 disfavors r1 . 0.1, unless the χ1 annihilation via an off-shell dark photon
is close to resonance. On the other hand, we have found that WDM constraints on χ1 provide com-
plementary constraints on the kinetic mixing parameter for r1 & 0.1. We emphasize that the WDM
constraints redeem the relaxed constraints on the self-scattering cross section of χ1 for r1 < 1. This
motivates the further study of structure-formation constraints on the mixed DM scenarios. Actually,
some observations, e.g., the flux anomaly of quadrupole lens systems [92–99], and the redshifted 21 cm
signal [100–107], provides severer constraints in the case of pure WDM compared to the constraints
we took in this paper. If one reanalyzes data from such probes in the case of mixed DM, we may
get stronger constraints on the warmness of χ1 and hence increase the synergy between the warmness
constraints from structure formation and terrestrial experiments.
Acknowledgments
The authors would like to thank Doojin Kim for fruitful discussions and comments. The work of
A.K. and H.K. is supported by IBS under the project code, IBS-R018-D1. A.K. also acknowledges
partial support from Grant-in-Aid for Scientific Research from the Ministry of Education, Culture,
Sports, Science, and Technology (MEXT), Japan, 18K13535 and 19H04609; from World Premier
International Research Center Initiative (WPI), MEXT, Japan; from Norwegian Financial Mechanism
for years 2014-2021, grant nr 2019/34/H/ST2/00707; and from National Science Centre, Poland, grant
DEC-2018/31/B/ST2/02283. J.C.P. acknowledges support from the National Research Foundation of
Korea (NRF-2019R1C1C1005073 and NRF-2021R1A4A2001897). S.S. acknowledges support from the
National Research Foundation of Korea (NRF-2020R1I1A3072747).
Appendix A: Boltzmann equations for the boosted DM (BDM)
In this appendix, we derive the evolution equations for boosted DM (BDM). One can find the number
density evolution equations in Eq. (A15) and Eq. (A17), and the temperature evolution equation in
Eq. (A27).
30
1. Thermal averaged quantities
We lay-out the identities that we will utilize in this appendix. The thermal averaged quantities for
particle with mass m with the Boltzmann distribution f eq = exp [−E/T ] are given as∫p
f eq =m3
2π2
(T
m
)K2 (m/T ) ≡ neq ,∫
p
E f eq =m4
2π2
(T
m
)[K1 (m/T ) + 3
(T
m
)K2 (m/T )
]≡ ρeq ,∫
p
|p|22m
f eq =3
2
m4
2π2
(T
m
)2
K3 (m/T ) ≡ Keq ,∫p
|p|23E
f eq =m4
2π2
(T
m
)2
K2 (m/T ) ≡ P eq ,
(A1)
where∫p≡∫d3p/(2π)3 and Kn are the modified Bessel function of the second kind. From the top,
each quantities represent number density (neq), energy density (ρeq), kinetic-energy density in the non-
relativistic limit (Keq), and pressure (P eq). We will often perform derivatives of the thermal averaged
quantities:
K ′n (x) = −1
2[Kn−1 (x) +Kn+1 (x)] . (A2)
It is also useful to note the following recurrence relation for integer n:
Kn−1 (x)−Kn+1 (x) = −2n
xKn (x) . (A3)
We are interested in an epoch where dark matter is non-relativistic. Therefore, it is useful to note the
asymptotic behavior of Kn(x):
limx→∞
Kn (x) = e−x[√
π
2x+O
(x−3/2
)]. (A4)
2. Interaction of DM and the corresponding collision terms
We consider a case where two DM particles, χ0 and χ1, which were initially in thermal equilibrium
with a thermal plasma, e.g., SM plasma. The heavy state χ0 has no direct couplings to SM, and
annihilates into the light state χ1. On the other hand, χ1 interacts with SM, e.g., through a dark
photon portal, and its relic abundance is determined by the annihilation into SM particles. The yield
of the DM particles are determined by the chemical freeze-out of the following processes:
χ0 (1) + χ0 (2)↔ χ1 (3) + χ1 (4) ,
χ1 (1) + χ1 (2)↔ φ (3) + φ (4) ,(A5)
where φ is some SM state, and the number indices will be used when defining the collisional integrals,
as will be shown shortly. For simplicity, we assume χ0, χ1, and φ to be real scalars. Processes above
accompany elastic scatterings from the crossing symmetry:
χ0 (1) + χ1 (2)↔ χ0 (3) + χ1 (4) ,
χ1 (1) + φ (2)↔ χ1 (3) + φ (4) .(A6)
31
Efficient processes of Eq. (A6) keep DM in kinetic equilibrium with the SM plasma (Tχ1= T ) during
(and after) their chemical freeze-out.
We are interested in a case where χ1 exhibits sizable self-scattering σself/m ∼ 1 cm2/g:
χ1 (1) + χ1 (2)→ χ1 (3) + χ1 (4) . (A7)
Efficient self-scattering keeps the distribution function of χ1 in the equilibrium form, i.e., fχ1=
exp [−(E − µ)/Tχ1]. In the presence of the efficient self-scattering, the excess kinetic energy of the
boosted χ1’s produced from the first process of Eq. (A5) would be efficiently re-distributed to the
other χ1 particles; this results in the self-heating epoch after the kinetic decoupling of χ1.
Now that we have introduced the interactions of DM, we present their corresponding Boltzmann
equations. The Boltzmann equation for χ0 is given as[∂
∂t+Hp1
∂
∂p1
]fχ0
=1
2E1(Cann,χ0
+ Cel,χ0) , (A8)
where Cann,χ0 and Cel,χ0 are the collisional integral for annihilation [the first process of Eq. (A5)] and
elastic scattering [the first process of Eq. (A6)] of χ0, respectively. Cann,χ0is given as
Cann,χ0[fχ0
(p1)] = 2
∫dΠ2dΠ3dΠ4 (2π)
4δ4 (p1 + p2 − p3 − p4)
× |Mχ0χ0→χ1χ1|2 [fχ1
(p3) fχ1(p4)− fχ0
(p1) fχ0(p2)] .
(A9)
Note that if two identical particles participate in initial/final state, additional factor of 1/2’s are
present compared to the case of non-identical particles in the initial/final state; this is to correct the
over-counting of the equivalent phase-space configurations. In Eq. (A9), we implicitly assume that the
phase space integrals dΠi’s are done over inequivalent configurations so that the integrations implicitly
take into account the 1/2 factor(s); hereafter, we assume this convention. Cel,χ0 is the collision term
for kinetic interactions that keeps Tχ0= Tχ1
, e.g., the first process of Eq. (A6). For our purpose,
instead of explicitly writing down Cel,χ0 , we will assume that kinetic interactions are efficient so that
Tχ0= Tχ1
during the freeze-out of χ0, and decouples afterwards so that Tχ0∝ 1/a2.
Similarly, the Boltzmann equation for χ1 is given as[∂
Integrating Eq. (A8) (Eq. (A10)) with respect to d3p1/(2π)3 would give the evolution equation for
number density of χ0 (χ1), and integrating with E1-weighting will give the evolution equation for
temperatures of χ0 (χ1). We will present the derivations in the following subsections.
3. Equations for number density
We now derive the number density equations for χ0 and χ1. Specifically, we are interested in the
chemical freeze-out of the DM particles. For χ0, we assume that χ0 follows the temperature of χ1,
i.e., we assume fχ0= exp [−(E − µ)/Tχ1
], during its chemical freeze-out. this could be due to the
efficient kinetic interactions of χ0 with χ1 represented by Cel,χ0 in Eq. (A8). While we do not specify
the interactions for simplicity, we instead assume that the kinetic interactions are efficient during the
chemical freeze-out of χ0. Integrating Eq. (A8) over d3p1/(2π)3, we find the evolution equation for
number density of χ0:
nχ0 + 3Hnχ0 =
∫p1
1
2E1Cann,χ0 [fχ0 (p1)]
= 2
∫ 4∏i=1
dΠi (2π)4δ4 (p1 + p2 − p3 − p4)
× |Mχ0χ0→χ1χ1|2 [fχ1
(p3) fχ1(p4)− fχ0
(p1) fχ0(p2)]
= −〈σ0vrel〉Tχ0
[n2χ0−〈σ0vrel〉Tχ1
〈σ0vrel〉Tχ0
(neqχ0
(Tχ1)
neqχ1 (Tχ1)
)2
n2χ1
]
= −〈σ0vrel〉Tχ1
[n2χ0−(neqχ0
(Tχ1)
neqχ1 (Tχ1
)
)2
n2χ1
],
(A15)
where (σ2vrel) annihilation cross section of χ0, and we have set Tχ0= Tχ1
in the last equality. In the
first equality, integration over Cel,χ0vanishes under the assumption that kinetic interactions conserve
number of χ0. 〈σvrel〉 denotes the thermal average of the cross section; we multiply the distribution
function of initial particle states to the cross section, and integrate over all possible phase space
configuration (this is in contrast to our convention of integration over dΠi’s in C’s, where we only
integrate over inequivalent phase space configuration):
〈σvrel〉T =1
neq1 (T )neq
2 (T )
∫d3p1d
3p2 (σvrel) feq1 (p1;T ) f eq
2 (p2;T ) . (A16)
For χ1, efficient self-scattering of χ1 keeps its distribution proportional to Boltzmann distribution,
i.e., fχ1= exp [−(E − µ)/Tχ1
]. Efficient elastic scattering with SM states, e.g., χ1φ → χ1φ, keep χ1
in kinetic equilibrium (Tχ1 = T ) with the SM plasma during its chemical freeze-out. From Eq. (A10),
33
we find the evolution equation for number density of χ1:
nχ1+ 3Hnχ1
=
∫p1
1
2E1(Cinv,χ0
[fχ1(p1)] + Cann,χ1
[fχ1(p1)])
= 〈σ0vrel〉Tχ1
[n2χ0−(neqχ0
(Tχ1)
neqχ1 (Tχ1
)
)2
n2χ1
]
− 〈σ1vrel〉Tχ1
[n2χ1− 〈σ1vrel〉T〈σ1vrel〉Tχ1
neq2χ1
(T )
]
= 〈σ0vrel〉T
[n2χ0−(neqχ0
(T )
neqχ1 (T )
)2
n2χ1
]− 〈σ1vrel〉T
[n2χ1− neq2
χ1(T )],
(A17)
where we have set Tχ1 = T in the last equality. In the first equality, Cel,χ1 and Cself do not contribute
to the number density equation since they conserve number of χ1.
4. Equations for temperature
In Appendix A 3, we derived the number density equations under the assumption that χ0 and χ1
share the same temperature with the SM plasma (Tχ0 = Tχ1 = T ); this may be realized around
the chemical freeze-outs of χ0 and χ1, due to their efficient kinetic interactions; χ0χ1 → χ0χ1, and
χ1φ → χ1φ. However, the kinetic interactions will eventually decouple as the Universe cools down,
and we would need to follow the temperature evolution of χ0 (Tχ0) and χ1 (Tχ1
), independent from
the SM plasma temperature (T ).
In the Boltzmann equation for χ0 [Eq. (A8)], the kinetic interactions of χ0 is represented by Cel,χ0 .
While we do not specify the interactions for simplicity, one inevitable contribution is the χ0χ1 → χ0χ1
process. This process may be efficient around the freeze-out of χ0 and keep Tχ0 = Tχ1 , but likely to
decouple at the similar time of the freeze-out of χ0. 14 Again, for simplicity of our analysis, we assume
that kinetic equilibrium is achieved between χ0 and χ1 during the freeze-out of χ0, and decouples
afterwards so that Tχ0 ∝ 1/a2.
For χ1, we have a kinetic interaction between χ1 and the SM plasma, χ1φ → χ1φ. Due to un-
suppressed number density of the light SM particle φ, the kinetic equilibrium is likely to be maintained
until long after the freeze-out of χ1. Until the kinetic decoupling, the temperature redshifts as Tχ1∝
1/a. Unlike χ0, the temperature of χ1 would not redshift like non-relativistic free-streaming particles
(∝ 1/a2) because boosted χ1 are constantly produced from χ0 annihilation into χ1. The excess kinetic
energy of the boosted χ1’s will be redistributed to the other χ1’s through efficient self-scattering,
heating the χ1 particles as a whole. To investigate the evolution of χ1 around the kinetic decoupling,
we derive the evolution equation for Tχ1. Starting from Eq. (A10), we integrate it with E1-weighting.
Let us perform the integration for the LHS of Eq. (A10):∫p1
E1
[∂
∂t+Hp1
∂
∂p1
]fχ1 = ρχ1 + 3H (ρχ1 + Pχ1)
= (nχ1 + 3Hnχ1) 〈Eχ1〉Tχ1+ nχ1
(Tχ1
T 2χ1
σ2E,Tχ1
+ 3HTχ1
),
(A18)
14 Kinetic equilibrium during χ0’s freeze-out may be realized for large mass difference between χ0 and χ1. Meanwhile,
for almost degenerate masses, kinetic equilibrium of χ0 may not be achieved, since χ0χ1 → χ0χ1 may not be efficient.
34
where in the second equality, we assumed that the distribution function of χ1 is fχ1= [nχ1
/neqχ1
(Tχ1)]f eq(Tχ1
);
this is a reasonable assumption if the self-scattering of χ1 is efficient. Using the identities presented in
Appendix A 1, it is straightforward to achieve the second equality where σ2E,Tχ1
=⟨E2χ1
⟩Tχ1
−〈Eχ1〉2Tχ1,
and
〈Eχ1〉Tχ1
/mχ1=
ρeqχ1
(Tχ1)
mχ1neqχ1 (Tχ1)
=3
xχ1
+K1 (xχ1
)
K2 (xχ1)→ 1 +
3
2xχ1
, (A19)
⟨E2χ1
⟩Tχ1
/m2χ1
= 1 +2Keq
χ1
mχ1neqχ1
= 1 +3
xχ1
K3 (xχ1)
K2 (xχ1)→ 1 +
3
xχ1
, (A20)
σ2E,Tχ1
→ 3
2T 2χ1, (A21)
where xχ1= mχ1
/Tχ1, and the RHS of the arrows denote the non-relativistic limits. Putting together
with the RHS of the E1-weighted integral of Eq. (A10), the overall Boltzmann equation for Tχ1 is
given as
Tχ1
T 2χ1
+3HTχ1
σ2E,Tχ1
=1
nχ1σ2E,Tχ1
∫p1
1
2E1
×[(E1 − 〈Eχ1〉Tχ1
)Cinv,χ0 + Cann,χ1+ E1Cel,χ1
].
(A22)
Let us have a look at the collisional integrals one by one. The E1-weighted integral of Cinv,χ0 can be
manipulated as∫p1
(E1 − 〈Eχ1〉Tχ1
) Cinv,χ0
2E1=
∫p1
(E1 + E2
2− 〈Eχ1
〉Tχ1
)Cinv,χ0
2E1,
=
∫p1
(E3 + E4
2− 〈Eχ1〉Tχ1
)Cinv,χ0
2E1, (A23)
= 〈∆E0σ0vrel〉Tχ0
[n2χ0−(neqχ0
(Tχ1)
neqχ1 (Tχ1)
)2 〈∆E0σ0vrel〉Tχ1
〈∆E0σ0vrel〉Tχ0
n2χ1
],
where we have defined ∆E0 = Eχ0 − 〈Eχ1〉Tχ1. In the second equality, we have used the property
of the 4-momentum conserving δ-function. A similar expression holds for the E1-weighted integral of
Cann,χ1.
The remaining is the E1-weighted integral of Cel,χ1 . For numerical convenience, we adopt the analytic
approximation for Cel,χ1[fχ1
] from Ref. [64]:
Cel,χ1[fχ1
] ' 2mχ1
∂
∂p1,i
[γχ1sm
(mχ1
T∂fχ1
∂p1,i+ p1,ifχ1
)], (A24)
where we took the non-relativistic limit of χ1, T is the temperature of φ, and γχ1sm is the momentum
transfer rate given as
γχ1sm '1
6mχ1T
∑s2
∫d3p2
(2π)3 f
eq2
∫ 0
−4|p2|2dt (−t) dσχ1φ→χ1φ
dtvrel . (A25)
The sum denotes the spin degrees of freedom of φ. Note that the approximation of Eq. (A24) is done
in the limit where the momentum transfer pχ1,1 − pχ1,3 is smaller than the typical DM momentum;
this is a reasonable approximation for non-relativistic χ1 scattering with a much lighter relativistic SM
particle φ. The E1-weighted integral of Cel,χ1is then approximated as∫
p1
E1Cel,χ1 [fχ1(p1)]
2E1' −3nχ1γχ1sm (Tχ1 − T ) . (A26)
35
Putting altogether the Eq. (A22), Eq. (A23), and Eq. (A26), the temperature evolution equation is
given as
Tχ1
T 2χ1
+3HTχ1
σ2E,Tχ1
' 1
nχ1σ2E,Tχ1
〈∆E0σ0vrel〉Tχ0
[n2χ0−(neqχ0
(Tχ1)
neqχ1 (Tχ1
)
)2 〈∆E0σ0vrel〉Tχ1
〈∆E0σ0vrel〉Tχ0
n2χ1
]
− 〈∆E1σ1vrel〉Tχ1
[n2χ1− 〈∆E1σ1vrel〉T〈∆E1σ1vrel〉Tχ1
neq2χ1
(T )
](A27)
− 3nχ1γχ1sm (Tχ1
− T )
,
where ∆E1 = Eχ1− 〈Eχ1
〉Tχ1. In general, Eq. (A27) (together with the evolution equation for Tχ0
that we have not specified) and the number density evolution equations [Eq. (A15) and Eq. (A17)]
form a coupled system of Boltzmann equations. However, if the kinetic decoupling of χ1 takes place
well after the freeze-out of DM, Eq. (A27) can be effectively considered to be decoupled from the
number density equations, while taking the freeze-out number density for nχi . In the case where the
kinetic decoupling of χ1 interferes with the chemical freeze-out, one has to follow the co-evolution of
temperature and number density; such a case is shown as yellow hatched region in Figure 8.
For most of the parameter space we consider in the main text, the kinetic decoupling of χ1 is well
separated from the freeze-out of χ1. Furthermore, the DM particles remain non-relativistic throughout
their evolution, i.e., the DM temperatures are negligible compared to their masses and the mass
difference δm = mχ0− mχ1
. In such case, the RHS of Eq. (A27) is further simplified, leading to
Eq. (20): inside the squared parentheses of the first line, the second term which represents the cooling
of χ1 through the inverse process χ1χ1 → χ0χ0, is negligible to the first term since it is a kinematically
forbidden process; the second line, which corresponds to the heating and cooling through the χ1χ1 ↔φφ process, is negligible compared to the first line.
Appendix B: Freeze-out of DM
In this appendix, we give semi-analytic estimations for final yield of DM. We find that our semi-
analytic estimations agrees reasonably well with numerical solutions. Semi-analytic understanding to
the numerical solutions will be useful when scanning the viable parameter space for the two-component
DM scenario. One can find the relic density estimations in Eq. (B14), Eq. (B19), and Eq. (B25).
We rewrite the number density equations for χ0 and χ1 [Eqs. (A15) and (A17)] in terms of the yield
Y = n/s:
dYχ0
dx= −λχ0(x)
x
[Y 2χ0−(Y eqχ0
(x)
Y eqχ1 (x)
)2
Y 2χ1
], (B1)
dYχ1
dx=λχ0
(x)
x
[Y 2χ0−(Y eqχ0
(x)
Y eqχ1 (x)
)2
Y 2χ1
]− λχ1
(x)
x
[Y 2χ1−(Y eqχ1
(x))2]
, (B2)
where x = mχ1/T , and we assumed that (χ0, χ1) and (χ1, φ) are in kinetic equilibrium, Tχ0
= Tχ1= T .
The λ(x)’s are given as
λχ0(x) =
s 〈σ0vrel〉TH
[1− 1
3
d ln g?S (x)
d lnx
], (B3)
λχ1(x) =
s 〈σ1vrel〉TH
[1− 1
3
d ln g?S (x)
d lnx
]. (B4)
36
Hereafter, we will consider g? and g?S as constants for simplicity.
– Abundance of χ0.
Before numerically solving the number density equations, let us take a semi-analytical approach.
First, let us discuss the chemical freeze-out of χ0. We focus on the case where the freeze-out of χ0
is well separated from (and prior to) that of χ1. In such case, Yχ1would follow Y eq
χ1(x) around the
freeze-out of χ0, and the final yield of χ0 will be virtually the same as the standard case of WIMP. Let
us recall the estimation of final yield (Y∞) for WIMP, as we will do a similar analysis when estimating
the yield for χ1. Around the freeze-out of χ0, Yχ1 ' Y eqχ1
(x) and Eq. (B1) is approximated as
dYχ0
dx= −λχ0 (x)
x
[Y 2χ0−(Y eqχ0
(x))2]
. (B5)
In the region where χ0 is near the chemical equilibrium, Yχ0can be written as a small deviation from
Y eqχ0
:
Yχ0(x) = Y eq
χ0(x) + ∆Yχ0
(x) . (B6)
In the lowest order in ∆Yχ0 , Eq. (B5) is written as
dY eqχ0
dx' −λχ0
(x)
x2∆Yχ0
Y eqχ0. (B7)
Since dY eqχ0/dx ∼ −(mχ0
/mχ1)Y eqχ0
(note again that x = mχ1/T ) when χ0 is non-relativistic, we find
∆Yχ0
Y eqχ0
(x) ' (mχ0/mχ1
)x
2λχ0(x)Y eq
χ0 (x)=
(mχ0/mχ1
)x
2
(Γ (= neq
χ0〈σ0vrel〉)H
∣∣∣∣x
)−1
, (B8)
which implies that the relative deviation grows exponentially with x, since Y eqχ0
(x) ∝ x3/2 exp[−x]. We
define the freeze-out point, xfo,0, as the point when the relative deviation of Yχ0from Y eq
χ0starts to
exceed unity:
∆Yχ0
Y eqχ0
(xfo,0) ' (mχ0/mχ1
)xfo,0
2
(Γ
H
∣∣∣∣xfo,0
)−1
= 1 , (B9)
where we see the familiar Gamow’s criterion with an extra factor of x−1 multiplied to the reaction
rate Γ. Note that for 〈σ0vrel〉T that achieves the abundance of χ0 that is similar to the observed DM
abundance, xfo,0 ∼ 20 (mχ1/mχ0
). Now, let us examine the region well after the freeze-out, x & xfo,0.
Since the growth of the relative deviation with respect to x is exponential, Y eqχ0
would be ignorable
compared to Yχ0:
dYχ0
dx' −λχ0
(x)
xY 2χ0. (B10)
Given the boundary condition at xfo,0, this is a separable equation that we can solve:
1
Yχ0(x)
=1
Yχ0(xfo,0)
+
∫ x
xfo,0
dx′λχ0 (x′)x′
,
=1
Yχ0(xfo,0)
+λχ0 (xfo,0)
n0 + 1
[1−
(xfo,0
x
)n0+1]
︸ ︷︷ ︸=1/Yann,χ0 (x;xfo,0)
,(B11)
37
where 〈σ0vrel〉 ∝ x−n0 . Let us define YWIMP,χi(x) for a notational convenience:
YWIMP,χi(x) =ni + 1
λχi(x). (B12)
We get the final yield of χ0 by taking x→∞ in Eq. (B11); note that in this limit, the second term in
the RHS, 1/Yann,χ0(x;xfo,0) ' 1/YWIMP,χ0(xfo,0), dominates over the first term, 1/Yχ0(xfo,2), since
Yχ0(xfo) ' (mχ0
/mχ1)xfo,0
λχ0(xfo,0)
= (mχ0/mχ1
)xfo,0 ×YWIMP,χ0
(xfo,0)
n0 + 1, (B13)
which can be deduced from the definition of freeze-out point in Eq. (B9). Then the final abundance
of χ0, Yχ0(∞), is given as
Yχ0(∞) ' YWIMP,χ0
(xfo,0) =n0 + 1
λχ0 (xfo,0). (B14)
– Abundance of χ1; the case of constant Yast. with Yast. < YWIMP,χ1(xfo,1).
The chemical freeze-out of χ1 could be very different from χ0, since annihilation of χ0 continuously
produce χ1. Around the freeze-out of χ1, χ0 already froze-out and we may approximate Eq. (B2) as
dYχ1
dx' −λχ1
(x)
x
[Y 2χ1−(Y eqχ1
(x))2 − Y 2
ast. (x)], (B15)
where Yast. is defined as
Yast. (x) =
√〈σ0vrel〉T〈σ1vrel〉T
Yχ0(∞) . (B16)
First, let us first focus on a case where Yast. is constant, e.g., annihilation of χ0 and χ1 are both
s-wave. We define the standard freeze-out point of χ1 as xfo,1 = mχ1/Tfo,1 as in Tfo,1 ∼ mχ1
/20, as
in Eq. (B9). If Yast. is smaller than YWIMP,χ1(xfo,1) ' 1/λχ1
(xfo,1), then Yast. is smaller than Yχ1(x)
throughout the freeze-out of χ1 and the final yield of χ1 would be just the standard estimation given
in Eq. (B14). The reasoning is the following: if Yast. < YWIMP,χ1(xfo,1), then Yast. < Y eq
χ1(xfo,1) where
xfo,1 is defined by the WIMP freeze-out condition Eq. (B9) for χ1. Therefore, around x = xfo,1, we
can drop the Y 2ast. term in the RHS of Eq. (B15). Then, the number density equation for χ0 is the
same with Eq. (B5).
– Abundance of χ1; the case of constant Yast. with Yast. > YWIMP,χ1(xfo,1).
If Yast. > YWIMP,χ1 , there would be a point in x where Yχ1(x) decreases to become Yχ1(x) = Yast..
After such moment, Yχ1will freeze at the value of Yast.. We can justify this statement by using the
similar procedure that we have done for the freeze-out of χ0. If the moment when Yχ1(x) = Yast. takes
places when χ1 is non-relativistic, soon after then Y eqχ1
(x) would become negligible in Eq. (B15) due
to the exponential suppression. Let us take of a small deviation of Yχ1(x) from Yast.:
Yχ1(x) = Yast. + ∆Yχ1(x) . (B17)
Then Eq. (B15) leads to
dYast.
dx' −λχ1
(x)
x2∆Yχ1Yast. . (B18)
38
Since we are thinking of a case where Yast. is constant, Eq. (B18) implies that ∆Yχ1(x) ' 0 afterwards.
Thus, Yast. is a final yield of χ1 in the case of Yast. > YWIMP,χ1(xfo,1). Putting our results together for
the cases of Yast. < YWIMP,χ1(xfo,1) and Yast. > YWIMP,χ1(xfo,1), we may write
Yχ1,∞ ' max [Yast., YWIMP,χ1(xfo,1)] , (B19)
where Yast. is constant. The discrepancy between Eq. (B19) and the numerical solutions to Eq. (B1)
and Eq. (B2) is within O(10)%; see Figure 1.
– Abundance of χ1; the case of decreasing Yast.(x).
The remaining is the case where Yast.(x) is not constant. Again, we estimate the final yield of χ1 while
assuming that Yχ1(x) is initially following Yast.(x); we will compare this yield with YWIMP,χ1
(xfo,1)
and take a larger one as the final yield of χ1.
Let us consider a case where Yast.(x) ∝ xnast. is decreasing, i.e., nast. < 0. If Yχ1(x) is initially
following Yast.(x), Yχ1(x) would be decreasing as with Yast.(x). But Yχ1(x) would not be following
Yast.(x) indefinitely, but eventually depart from Yast.(x). From Eq. (B18), the relative deviation of
Yχ1(x) from Yast.(x) becomes order unity when x = x′fo, i.e., ∆Yχ1(x)/Yast.(x) = c′ where c′ is a
O(1) constant for fitting our analytic estimations on final relic abundance with numerical results. For
general values of nast., the deviation point x′fo is determined by the condition given as
|nast.|2λχ1
(x′fo)Yast. (x′fo)= c′ , (B20)
where nast. = (n1 − n2)/2. Afterwards, Eq. (B15) becomes
dYχ1
dx' −λχ1
(x)
xY 2χ1, (B21)
and the solution for Yχ1 is given as
1
Yχ1(x)
=1
Yχ1(x′fo)
+λχ1 (x′fo)
n1 + 1
[1−
(x′fox
)n1+1]. (B22)
with 〈σ1vrel〉T ∝ x−n1 . Taking x→∞, the second term in the RHS is similar to the standard case for
freeze-out [Eq. (B14)] while λχ1is evaluated at x′fo, i.e., 1/YWIMP,χ1
(x′fo). Meanwhile, we can see that
the first term and the second term in the RHS of Eq. (B22) gives similar contribution to the final yield of
χ1; they are both of order ∼ 1/YWIMP,χ1(x′fo). This is different from the standard freeze-out of WIMP;
in Eq. (B11), the first term in the RHS is negligible compared to the second term. The difference
stems from the fact that Y eqχ0
(x) decreases exponentially, while Yast.(x) decreases by a power-law. In
the LHS of Eq. (B7), since Y eqχ0
(x) decreases exponentially, we had dY eqχ0/dx ' −Y eq
χ0. On the contrary,
in the LHS of Eq. (B18), since Yast.(x) decreases by a power-law, we had dYast./dx = (nast./x)Yast..
The additional factor of 1/x in the LHS of Eq. (B18) makes the difference from the case of WIMP.
From this observation, we will simply estimate the final yield of χ1 as YWIMP,χ1(x′fo). The left panel
of Figure 10 shows the comparison between [Eq. (B22)] (horizontal blue) and the numerical solution
of Yχ1(solid blue); we have set c′ ' 0.63 to match our estimation YWIMP,χ1
(x′fo) with the numerical
solutions. With the calibration of c′ at hand, we estimate the required annihilation cross section of
χ1 by requiring YWIMP,χ1(x′fo) to be equal to the desired final yield of χ1. Note that we may also use
Eq. (B22) to estimate the final yield of χ1, while we would arrive at a different value of c′; nevertheless,
39
we would end up with the same required annihilation cross section of χ1.
– Abundance of χ1; the case of increasing Yast.(x).
Similarly, in the case where Yast.(x) is increasing (nast. > 0), the departure point x′fo of Yχ1(x) from
Yast.(x) is determined by Eq. (B20). Afterward the departure point, Eq. (B15) becomes
dYχ1
dx' λχ1
(x)
xY 2
ast.(x) , (B23)
and the solution is
Yχ1(x) = Yχ1
(x′fo) +λχ1
(x′fo)Y 2ast. (x
′fo)
n0 + 1
[1−
(x′fox
)n0+1]. (B24)
Taking x → ∞, the second term in the RHS is similar to the case of WIMP [Eq. (B14)] while λχ1
is evaluated at x′fo. We see again that the first term and the second term in the RHS of Eq. (B24)
have comparable contribution to the final yield of χ1, i.e., Yast.(x′fo) ∼ Yχ1
(x′fo) ∼ 1/λχ1(x′fo): we
again simply estimate the final yield of χ1 as YWIMP,χ1(x′fo). The left panel of Figure 2 shows the
comparison between our estimation YWIMP,χ1(x′fo) (horizontal blue) and the numerical solution of Yχ1
(solid blue); we have set c′ ' 0.35 to match our estimation YWIMP,χ1(x′fo) with the numerical solutions.
In both of the cases when Yast.(x) is decreasing/increasing, we see that if Yχ1(x) was initially following
Yast.(x), YWIMP,χ1(x′fo) ' 1/λχ1
(x′fo) is a reasonable estimate for the final yield of χ1. It is natural to
ask what is the precise condition for Yχ1(x) to initially follow Y12(x) to happen, and how to determine
the final yield of χ1 when Yχ1(x) is not initially following Yast.(x). This is a question that is difficult
to study analytically, and we would need to depend on numerical analyses. Nevertheless, we give the
following estimation for final yield of χ1 when Yast.(x) is time-dependent:
Yχ1,∞ ' max [YWIMP,χ1(x′fo) , YWIMP,χ1
(xfo,1)] , (B25)
where we note again that x′fo (xfo,1) is defined by Eq. (B20) (Eq. (B9)). Although somewhat crude,
Eq. (B25) physically makes sense. For example, if we think of a limit where Yast.(x) → 0, x′foalso decreases according to Eq. (B20), since λχ1
(x) is generally a decreasing function of x. Then,
YWIMP,χ1(x′fo)→ 0 by the same reasoning, and the final yield of χ1 is just the standard WIMP yield;
this should be since the Yast.(x)→ 0 limit corresponds to the standard freeze-out.
Appendix C: Temperature evolution of χ1 in the case of p-wave annihilation of χ0
In this appendix, we comment on the temperature evolution of χ1 when χ0-annihilation is p-wave.
Let us examine the asymptotic behavior of Tχ1from the evolution equation Eq. (20). We take an
ansatz that Tχ1 ∝ 1/aN , where N ≤ 2 is some positive number. As in the main text, we denote the
non-relativistic limit of 〈σ0vrel〉Tχ0as (σ0vrel). The ansatz would mean Tχ1
+NHTχ1= 0. Furthermore,
since the kinetic decoupling of χ1 takes place after the freeze-out of χ0 and χ1, their yields Yχi = nχi/s
are virtually conserved. For our ansatz to be consistent with Eq. (20), following condition must hold:
<latexit sha1_base64="bY2eIjCrpLyCsTuR28am3F0cHVg=">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</latexit> r 1(=